Release v26.04.16.1 — Fix tool actions mismatch in system prompt

fix(llm): sync available actions list with actual registered tools
The system prompt listed phantom tools (create_task, delete_task, get_note) that don't exist, causing the model to spiral when users asked to create tasks under a project. Replaced the stale hardcoded string with a dynamically-built actions list matching all registered tools, and added conditional searxng/caldav extensions. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-04-16 12:39:02 +00:00 · 2026-04-16 08:16:36 -04:00 · 2026-04-16 05:52:47 +00:00 · 2026-04-15 23:29:14 -04:00 · 2026-04-15 22:35:39 -04:00 · 2026-04-15 21:09:16 -04:00
20 changed files with 1696 additions and 704 deletions
@@ -1,8 +1,12 @@
 # CI runs first; build only proceeds if all checks pass.
 #
-# Push to any branch: typecheck + lint + test
-# Push to dev:        gates + build :dev + :<sha>
-# Tag v* (release):   gates + build :latest + :<sha> + :<version>
+# Push to dev:       typecheck + lint + test + build :dev + :<sha>
+# Tag v* (release):  typecheck + lint + test + build :latest + :<sha> + :<version>
+#
+# main pushes are NOT gated here: a merge to main only happens after
+# dev has already passed CI, and the release tag is the sole trigger
+# for a production image. Re-running CI on the merge commit just burns
+# runner time without changing the outcome.
 #
 # To cut a release:
 #   Create a release via the Forgejo UI on main with a v* tag name.
@@ -11,6 +15,13 @@
 # PRs aren't triggered on purpose — this is a solo dev→main flow, so
 # gating on branch push is already enough.
 #
+# NOTE on the `if:` guards below: Forgejo Actions does not consistently
+# honor `on.push.branches` as a filter — merge commits landing on main
+# still trigger the workflow, producing redundant runs on the same SHA
+# that was already gated on dev. Every job therefore repeats the ref
+# check so main pushes trigger the workflow but every job skips
+# immediately (no runner time, no duplicate work).
+#
 # Required secrets (repo → Settings → Secrets → Actions):
 #   REGISTRY_USER   — your Forgejo username
 #   REGISTRY_TOKEN  — Forgejo PAT with write:packages scope
@@ -18,7 +29,7 @@ name: CI & Build

 on:
  push:
-    branches: [dev, main]
+    branches: [dev]
    tags: ["v*"]
    paths:
      - "src/**"
@@ -51,6 +62,8 @@ env:
 jobs:
  typecheck:
    name: TypeScript typecheck
+    # Skip on main merge-commit pushes — see workflow header comment.
+    if: github.ref == 'refs/heads/dev' || startsWith(github.ref, 'refs/tags/v')
    runs-on: ci-runner
    steps:
      - uses: actions/checkout@v6
@@ -77,6 +90,7 @@ jobs:

  lint:
    name: Python lint
+    if: github.ref == 'refs/heads/dev' || startsWith(github.ref, 'refs/tags/v')
    runs-on: ci-runner
    steps:
      - uses: actions/checkout@v6
@@ -88,6 +102,7 @@ jobs:

  test:
    name: Python tests
+    if: github.ref == 'refs/heads/dev' || startsWith(github.ref, 'refs/tags/v')
    runs-on: ci-runner
    steps:
      - uses: actions/checkout@v6
@@ -118,11 +133,10 @@ jobs:
    name: Build & push image
    needs: [typecheck, lint, test]
    # Build on dev branch pushes and version tag pushes only.
-    # main branch pushes run the gates for safety but do not build —
-    # the release tag (v*) is the sole trigger for a production image.
-    if: |
-      github.event_name == 'push' &&
-      (github.ref == 'refs/heads/dev' || startsWith(github.ref, 'refs/tags/'))
+    # Mirrors the ref guard on the gate jobs above — main merge-commit
+    # pushes skip here too, so no production image is ever built from a
+    # raw main push (only from the v* tag the release creates).
+    if: github.ref == 'refs/heads/dev' || startsWith(github.ref, 'refs/tags/v')
    runs-on: ci-runner
    permissions:
      contents: read
@@ -0,0 +1,34 @@
+"""Add content_full and context_prepared caches to rss_items.
+
+Revision ID: 0038
+Revises: 0037
+Create Date: 2026-04-13
+"""
+
+from __future__ import annotations
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = "0038"
+down_revision = "0037"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column("rss_items", sa.Column("content_full", sa.Text(), nullable=True))
+    op.add_column(
+        "rss_items",
+        sa.Column("context_prepared", sa.Text(), nullable=True),
+    )
+    op.add_column(
+        "rss_items",
+        sa.Column("content_fetched_at", sa.DateTime(timezone=True), nullable=True),
+    )
+
+
+def downgrade() -> None:
+    op.drop_column("rss_items", "content_fetched_at")
+    op.drop_column("rss_items", "context_prepared")
+    op.drop_column("rss_items", "content_full")
@@ -0,0 +1,38 @@
+"""Link rss_items to their discussion-summary note.
+
+Revision ID: 0039
+Revises: 0038
+Create Date: 2026-04-14
+"""
+
+from __future__ import annotations
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = "0039"
+down_revision = "0038"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "rss_items",
+        sa.Column("discussion_note_id", sa.BigInteger(), nullable=True),
+    )
+    op.create_foreign_key(
+        "fk_rss_items_discussion_note_id",
+        "rss_items",
+        "notes",
+        ["discussion_note_id"],
+        ["id"],
+        ondelete="SET NULL",
+    )
+
+
+def downgrade() -> None:
+    op.drop_constraint(
+        "fk_rss_items_discussion_note_id", "rss_items", type_="foreignkey"
+    )
+    op.drop_column("rss_items", "discussion_note_id")
@@ -313,7 +313,7 @@ See [sso-oauth.md](sso-oauth.md) for provider-specific setup instructions.

 ### Tool Routing

-No separate intent router — the main model handles all tool routing directly via Ollama's structured tool-calling output. The model receives the full tool schema list and decides whether to call a tool or respond conversationally. A thinking-mode heuristic (`_should_think()`) detects complex prompts and enables extended reasoning.
+No separate intent router — the main model handles all tool routing directly via Ollama's structured tool-calling output. The model receives the full tool schema list and decides whether to call a tool or respond conversationally. Extended reasoning (`think=True`) is always on for qwen3-class models: content-based gating was tried but exposed tool-call template fragility on short tool-intent prompts.

 ### Tool Loop

@@ -426,7 +426,9 @@ export async function deleteRssReaction(rssItemId: number): Promise<void> {
  return apiDelete(`/api/briefing/rss-reactions/${rssItemId}`);
 }

-export async function openArticleInChat(itemId: number): Promise<{ conversation_id: number }> {
+export async function openArticleInChat(
+  itemId: number,
+): Promise<{ conversation_id: number; assistant_message_id?: number; status?: string }> {
  return apiPost(`/api/chat/from-article/${itemId}`, {});
 }

@@ -54,6 +54,8 @@
  --page-max-width: 1200px;
  --page-padding-x: 1rem;
  --sidebar-width: 260px;
+  --chat-reading-width: min(1200px, 100%);
+  --chat-context-sidebar-width: 220px;
  --color-accent-warm: #b8860b;
  --color-accent-warm-light: #d4a017;
  --color-primary-solid: #7c3aed;
@@ -1,8 +1,11 @@
 <script setup lang="ts">
-import { ref, computed, nextTick } from 'vue'
+import { ref, computed, nextTick, onMounted, onUnmounted } from 'vue'
 import { apiGet, transcribeAudio } from '@/api/client'
 import { useVoiceRecorder } from '@/composables/useVoiceRecorder'
 import { useSilenceDetector } from '@/composables/useSilenceDetector'
+import { useStreamingTts } from '@/composables/useStreamingTts'
+import { useVoiceAudio, setVoiceVolume } from '@/composables/useVoiceAudio'
+import { useListenMode } from '@/composables/useListenMode'
 import { useChatStore } from '@/stores/chat'
 import { useSettingsStore } from '@/stores/settings'
 import { useToastStore } from '@/stores/toast'
@@ -29,6 +32,32 @@ const emit = defineEmits<{
 const store = useChatStore()
 const settingsStore = useSettingsStore()
 const voiceEnabled = computed(() => settingsStore.voiceSttReady)
+const voiceTtsEnabled = computed(() => settingsStore.voiceTtsReady)
+
+// ── Streaming TTS (listen mode) ───────────────────────────────────────────────
+const listenMode = useListenMode()
+const audio = useVoiceAudio()
+const speakerPopoverOpen = ref(false)
+
+const tts = useStreamingTts({
+  streamingContent: computed(() => store.streamingContent),
+  streaming: computed(() => store.streaming),
+  enabled: computed(() => listenMode.value && voiceTtsEnabled.value),
+})
+
+function lastAssistantContent(): string {
+  const msgs = store.currentConversation?.messages ?? []
+  return [...msgs].reverse().find((m) => m.role === 'assistant')?.content ?? ''
+}
+
+function toggleListen() {
+  listenMode.value = !listenMode.value
+  if (listenMode.value) {
+    tts.speak(lastAssistantContent())
+  } else {
+    tts.stop()
+  }
+}

 // ── Core input ────────────────────────────────────────────────────────────────
 const messageInput = ref('')
@@ -114,6 +143,20 @@ const transcribingVoice = ref(false)
 const recorder = useVoiceRecorder()
 const silenceDetector = useSilenceDetector()

+// Live mic halo. A solid red disc sits behind the mic button and scales
+// with `silenceDetector.amplitude` (0..1 RMS, already amplified for
+// visibility). The button itself stays put so the mic icon is always
+// legible on top. 0.2 baseline breathes on silence; loud speech drives
+// the disc out to ~2.6× the button size with a softer radial fade.
+const micGlowStyle = computed(() => {
+  if (!recorder.recording.value) return { display: 'none' }
+  const pulse = 0.2 + Math.min(silenceDetector.amplitude.value, 1) * 0.8
+  return {
+    transform: `translate(-50%, -50%) scale(${1 + pulse * 1.6})`,
+    opacity: String(0.45 + pulse * 0.4),
+  }
+})
+
 async function toggleVoice() {
  if (transcribingVoice.value) return
  if (recorder.recording.value) {
@@ -159,6 +202,15 @@ async function stopRecording() {
  finally { transcribingVoice.value = false }
 }

+// ── Click-outside to dismiss speaker popover ──────────────────────────────────
+function onDocumentMousedown(e: MouseEvent) {
+  if (!speakerPopoverOpen.value) return
+  const target = e.target as HTMLElement | null
+  if (!target?.closest('.speaker-wrapper')) speakerPopoverOpen.value = false
+}
+onMounted(() => document.addEventListener('mousedown', onDocumentMousedown))
+onUnmounted(() => document.removeEventListener('mousedown', onDocumentMousedown))
+
 // ── Exposed interface ─────────────────────────────────────────────────────────
 function focus() {
  inputEl.value?.focus()
@@ -230,23 +282,72 @@ defineExpose({ focus, prefill })
        class="input-textarea"
      ></textarea>

-      <!-- PTT mic -->
-      <button
-        v-if="voiceEnabled"
-        class="btn-icon btn-mic"
-        :class="{ 'mic-recording': recorder.recording.value, 'mic-transcribing': transcribingVoice }"
-        @click.prevent="toggleVoice"
-        :disabled="transcribingVoice || !store.chatReady"
-        :title="recorder.recording.value ? 'Click to stop (or wait for silence)' : 'Click to speak'"
-        :aria-label="recorder.recording.value ? 'Stop recording' : 'Start recording'"
-      >
+      <!-- Speaker / listen-mode popover -->
+      <div v-if="voiceTtsEnabled" class="speaker-wrapper">
+        <button
+          class="btn-icon btn-speaker"
+          :class="{ 'speaker-active': listenMode, 'speaker-busy': tts.speaking.value }"
+          @click="speakerPopoverOpen = !speakerPopoverOpen"
+          :title="listenMode ? 'Listen mode on' : 'Listen / volume'"
+          aria-label="Listen and volume settings"
+        >
+          <svg v-if="!tts.speaking.value" width="17" height="17" viewBox="0 0 24 24" fill="currentColor">
+            <path d="M3 9v6h4l5 5V4L7 9H3zm13.5 3c0-1.77-1.02-3.29-2.5-4.03v8.05c1.48-.73 2.5-2.25 2.5-4.02zM14 3.23v2.06c2.89.86 5 3.54 5 6.71s-2.11 5.85-5 6.71v2.06c4.01-.91 7-4.49 7-8.77s-2.99-7.86-7-8.77z"/>
+          </svg>
+          <svg v-else width="17" height="17" viewBox="0 0 24 24" fill="currentColor">
+            <path d="M18 12c0-1.77-1.02-3.29-2.5-4.03v2.21l2.45 2.45c.03-.2.05-.41.05-.63zm2.5 0c0 .94-.2 1.82-.54 2.64l1.51 1.51C21.8 14.82 22 13.43 22 12c0-4.28-2.99-7.86-7-8.77v2.06c2.89.86 5 3.54 5 6.71zM4.27 3L3 4.27 7.73 9H3v6h4l5 5v-6.73l4.25 4.25c-.67.52-1.42.93-2.25 1.18v2.06c1.38-.31 2.63-.95 3.69-1.81L19.73 21 21 19.73l-9-9L4.27 3zM12 4L9.91 6.09 12 8.18V4z"/>
+          </svg>
+        </button>
+        <div v-if="speakerPopoverOpen" class="speaker-popover">
+          <button
+            class="speaker-row speaker-row--toggle"
+            :class="{ 'speaker-row--active': listenMode }"
+            @click="toggleListen"
+          >
+            <span class="speaker-row-label">{{ listenMode ? 'Listening' : 'Read aloud' }}</span>
+            <span class="speaker-switch" :class="{ on: listenMode }"></span>
+          </button>
+          <div class="speaker-row">
+            <span class="speaker-row-label">Volume</span>
+            <input
+              type="range" min="0" max="1" step="0.05"
+              :value="audio.volume.value"
+              @input="setVoiceVolume(+($event.target as HTMLInputElement).value)"
+              class="speaker-volume-range"
+              aria-label="Volume"
+            />
+            <span class="speaker-volume-pct">{{ Math.round(audio.volume.value * 100) }}%</span>
+          </div>
+          <button
+            v-if="tts.speaking.value"
+            class="speaker-row speaker-row--stop"
+            @click="tts.stop()"
+          >
+            <svg width="12" height="12" viewBox="0 0 14 14" fill="currentColor"><rect width="14" height="14" rx="2"/></svg>
+            <span>Stop playback</span>
+          </button>
+        </div>
+      </div>
+
+      <!-- PTT mic (with live amplitude halo behind) -->
+      <div v-if="voiceEnabled" class="mic-wrapper">
+        <div class="mic-glow" :style="micGlowStyle" aria-hidden="true"></div>
+        <button
+          class="btn-icon btn-mic"
+          :class="{ 'mic-recording': recorder.recording.value, 'mic-transcribing': transcribingVoice }"
+          @click.prevent="toggleVoice"
+          :disabled="transcribingVoice || !store.chatReady"
+          :title="recorder.recording.value ? 'Click to stop (or wait for silence)' : 'Click to speak'"
+          :aria-label="recorder.recording.value ? 'Stop recording' : 'Start recording'"
+        >
        <svg v-if="!transcribingVoice" width="17" height="17" viewBox="0 0 24 24" fill="currentColor">
          <path d="M12 14c1.66 0 3-1.34 3-3V5c0-1.66-1.34-3-3-3S9 3.34 9 5v6c0 1.66 1.34 3 3 3zm-1-9c0-.55.45-1 1-1s1 .45 1 1v6c0 .55-.45 1-1 1s-1-.45-1-1V5zm6 6c0 2.76-2.24 5-5 5s-5-2.24-5-5H5c0 3.53 2.61 6.43 6 6.92V21h2v-3.08c3.39-.49 6-3.39 6-6.92h-2z"/>
        </svg>
        <svg v-else width="17" height="17" viewBox="0 0 24 24" fill="currentColor">
          <circle cx="12" cy="12" r="8" opacity="0.35"/><circle cx="12" cy="12" r="4"/>
        </svg>
-      </button>
+        </button>
+      </div>

      <!-- Abort (streaming) or Send -->
      <button
@@ -343,9 +444,50 @@ defineExpose({ focus, prefill })
 .btn-icon:hover { opacity: 1; }
 .btn-icon:disabled { opacity: 0.3; cursor: default; }

-.btn-mic.mic-recording { opacity: 1; color: #ef4444; }
+.btn-mic.mic-recording {
+  opacity: 1;
+  /* White icon sits on top of the red halo so it stays legible at any
+     pulse size. */
+  color: #fff;
+  position: relative;
+  z-index: 1;
+}
 .btn-mic.mic-transcribing { opacity: 0.5; }

+/* Mic wrapper + live amplitude halo. The glow is a filled red disc
+   absolutely positioned behind the button, scaled/faded by the
+   computed `micGlowStyle` so the user gets unmistakable feedback
+   that their voice is being picked up while the mic icon itself
+   stays put and readable on top. */
+.mic-wrapper {
+  position: relative;
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  flex-shrink: 0;
+}
+.mic-glow {
+  position: absolute;
+  top: 50%;
+  left: 50%;
+  width: 28px;
+  height: 28px;
+  border-radius: 50%;
+  background: radial-gradient(
+    circle,
+    rgba(239, 68, 68, 0.9) 0%,
+    rgba(239, 68, 68, 0.6) 55%,
+    rgba(239, 68, 68, 0) 100%
+  );
+  transform: translate(-50%, -50%) scale(1);
+  transform-origin: center;
+  pointer-events: none;
+  /* Smooth between amplitude samples (~100ms) so the pulse feels continuous
+     rather than stepping. */
+  transition: transform 0.12s ease-out, opacity 0.12s ease-out;
+  z-index: 0;
+}
+
 .note-picker-wrapper { position: relative; }
 .note-picker-dropdown {
  position: absolute;
@@ -418,4 +560,77 @@ defineExpose({ focus, prefill })
  flex-shrink: 0;
 }
 .btn-abort-inline:hover { border-color: #ef4444; color: #ef4444; }
+
+/* Speaker / listen-mode popover */
+.speaker-wrapper { position: relative; display: inline-flex; flex-shrink: 0; }
+.btn-speaker.speaker-active { opacity: 1; color: var(--color-primary); }
+.btn-speaker.speaker-busy { opacity: 1; color: #f59e0b; }
+.speaker-popover {
+  position: absolute;
+  bottom: calc(100% + 8px);
+  right: 0;
+  min-width: 220px;
+  background: var(--color-bg-card);
+  border: 1px solid var(--color-border);
+  border-radius: var(--radius-md);
+  box-shadow: 0 4px 16px var(--color-shadow);
+  padding: 0.35rem;
+  z-index: 30;
+  display: flex;
+  flex-direction: column;
+  gap: 0.2rem;
+}
+.speaker-row {
+  display: flex;
+  align-items: center;
+  gap: 0.6rem;
+  padding: 0.45rem 0.6rem;
+  background: none;
+  border: none;
+  border-radius: var(--radius-sm);
+  color: var(--color-text);
+  font-size: 0.85rem;
+  cursor: default;
+  text-align: left;
+  width: 100%;
+}
+.speaker-row--toggle { cursor: pointer; }
+.speaker-row--toggle:hover { background: var(--color-bg-secondary); }
+.speaker-row--active { color: var(--color-primary); }
+.speaker-row--stop {
+  cursor: pointer;
+  color: var(--color-text-muted);
+  justify-content: flex-start;
+}
+.speaker-row--stop:hover { color: #ef4444; background: var(--color-bg-secondary); }
+.speaker-row-label { flex: 1; }
+.speaker-volume-range { flex: 1; min-width: 0; }
+.speaker-volume-pct {
+  font-size: 0.75rem;
+  color: var(--color-text-muted);
+  min-width: 2.5rem;
+  text-align: right;
+}
+.speaker-switch {
+  position: relative;
+  width: 28px;
+  height: 16px;
+  border-radius: 9999px;
+  background: var(--color-border);
+  transition: background 0.15s;
+  flex-shrink: 0;
+}
+.speaker-switch::after {
+  content: '';
+  position: absolute;
+  top: 2px;
+  left: 2px;
+  width: 12px;
+  height: 12px;
+  border-radius: 50%;
+  background: var(--color-bg-card);
+  transition: transform 0.15s;
+}
+.speaker-switch.on { background: var(--color-primary); }
+.speaker-switch.on::after { transform: translateX(12px); }
 </style>
@@ -1,11 +1,6 @@
 <script setup lang="ts">
-import { ref, computed, watch, nextTick, onMounted } from 'vue'
+import { ref, computed, watch, nextTick, onMounted, onUnmounted } from 'vue'
 import { useChatStore } from '@/stores/chat'
-import { useSettingsStore } from '@/stores/settings'
-import { useStreamingTts } from '@/composables/useStreamingTts'
-import { useVoiceAudio, setVoiceVolume } from '@/composables/useVoiceAudio'
-import { useListenMode } from '@/composables/useListenMode'
-import { apiGet } from '@/api/client'
 import ChatMessage from '@/components/ChatMessage.vue'
 import ChatStreamingBubble from '@/components/ChatStreamingBubble.vue'
 import ChatInputBar from '@/components/ChatInputBar.vue'
@@ -34,33 +29,6 @@ const emit = defineEmits<{
 }>()

 const store = useChatStore()
-const settingsStore = useSettingsStore()
-
-// ── Voice / TTS ───────────────────────────────────────────────────────────────
-const voiceTtsEnabled = computed(() => settingsStore.voiceTtsReady)
-const listenMode = useListenMode()
-const audio = useVoiceAudio()
-const showVolumeSlider = ref(false)
-
-const tts = useStreamingTts({
-  streamingContent: computed(() => store.streamingContent),
-  streaming: computed(() => store.streaming),
-  enabled: computed(() => listenMode.value && voiceTtsEnabled.value),
-})
-
-function lastAssistantContent(): string {
-  const msgs = store.currentConversation?.messages ?? []
-  return [...msgs].reverse().find((m) => m.role === 'assistant')?.content ?? ''
-}
-
-function toggleListen() {
-  listenMode.value = !listenMode.value
-  if (listenMode.value) {
-    tts.speak(lastAssistantContent())
-  } else {
-    tts.stop()
-  }
-}

 // ── Scroll ────────────────────────────────────────────────────────────────────
 const messagesEl = ref<HTMLElement | null>(null)
@@ -89,47 +57,6 @@ watch(() => store.streamingToolCalls, (calls) => {
 }, { deep: true })
 watch(() => store.streaming, (s) => { if (!s) _seenCalendarToolIdx.clear() })

-// ── RAG scope chip (full, non-briefing, non-workspace) ────────────────────────
-const projects = ref<{ id: number; title: string }[]>([])
-const scopeDropdownOpen = ref(false)
-const scopePulse = ref(false)
-
-const showScopeChip = computed(
-  () => props.variant === 'full' && !props.briefingMode && !props.projectId
-)
-
-const scopeLabel = computed(() => {
-  const id = store.ragProjectId
-  if (id === -1) return 'All notes'
-  if (id === null) return 'Orphan notes'
-  return projects.value.find((p) => p.id === id)?.title ?? `Project ${id}`
-})
-
-async function loadProjects() {
-  try {
-    const data = await apiGet<{ projects: { id: number; title: string }[] }>('/api/projects?status=active')
-    projects.value = data.projects ?? []
-  } catch {
-    projects.value = []
-  }
-}
-
-async function onScopeSelect(value: number | null) {
-  scopeDropdownOpen.value = false
-  const convId = store.currentConversation?.id
-  if (!convId) return
-  await store.updateRagScope(convId, value)
-  scopePulse.value = true
-  setTimeout(() => { scopePulse.value = false }, 600)
-}
-
-// Pulse when model-driven scope change fires
-watch(() => store.ragProjectId, () => {
-  if (!showScopeChip.value) return
-  scopePulse.value = true
-  setTimeout(() => { scopePulse.value = false }, 600)
-})
-
 // ── Note context (full variant — included / suggested / auto-injected notes) ──
 const includedNoteIds = ref<Set<number>>(new Set())
 const includedNotes = ref<{ id: number; title: string }[]>([])
@@ -190,11 +117,92 @@ function removeIncludedNote(noteId: number) {
  }
 }

-const hasContextSidebar = computed(
-  () => props.variant === 'full' && !props.briefingMode && !props.projectId &&
-    (autoInjectedNotes.value.length > 0 || includedNotes.value.length > 0 || suggestedNotes.value.length > 0)
+const contextCount = computed(
+  () => autoInjectedNotes.value.length + suggestedNotes.value.length + includedNotes.value.length
 )

+const hasContextData = computed(
+  () => props.variant === 'full' && !props.briefingMode && !props.projectId && contextCount.value > 0
+)
+
+// ── Narrow-viewport sidebar toggle ────────────────────────────────────────────
+const NARROW_BREAKPOINT = 1200
+const isNarrow = ref(typeof window !== 'undefined' && window.innerWidth < NARROW_BREAKPOINT)
+const sidebarOpen = ref(false)
+
+function onResize() {
+  isNarrow.value = window.innerWidth < NARROW_BREAKPOINT
+}
+onMounted(() => window.addEventListener('resize', onResize))
+onUnmounted(() => window.removeEventListener('resize', onResize))
+
+function toggleContextSidebar() {
+  sidebarOpen.value = !sidebarOpen.value
+}
+
+const hasContextSidebar = computed(() => hasContextData.value && sidebarOpen.value)
+
+// ── Collapsible sections (per-conversation, localStorage) ─────────────────────
+type SectionKey = 'auto' | 'suggested' | 'included'
+const collapsedSections = ref<Set<SectionKey>>(new Set())
+
+function storageKey(): string | null {
+  const id = store.currentConversation?.id
+  return id ? `fa_chat_ctx_collapsed_${id}` : null
+}
+
+function loadCollapsed() {
+  const key = storageKey()
+  if (!key) { collapsedSections.value = new Set(); return }
+  try {
+    const raw = localStorage.getItem(key)
+    if (!raw) { collapsedSections.value = new Set(); return }
+    const arr = JSON.parse(raw) as SectionKey[]
+    collapsedSections.value = new Set(arr)
+  } catch {
+    collapsedSections.value = new Set()
+  }
+}
+
+function saveCollapsed() {
+  const key = storageKey()
+  if (!key) return
+  try {
+    localStorage.setItem(key, JSON.stringify([...collapsedSections.value]))
+  } catch { /* ignore */ }
+}
+
+function toggleSection(key: SectionKey) {
+  const next = new Set(collapsedSections.value)
+  if (next.has(key)) next.delete(key)
+  else next.add(key)
+  collapsedSections.value = next
+  saveCollapsed()
+}
+
+watch(() => store.currentConversation?.id, () => loadCollapsed(), { immediate: true })
+
+// ── Empty-state (full variant, fresh/empty conversation) ─────────────────────
+const showEmptyState = computed(
+  () =>
+    props.variant === 'full'
+    && !props.briefingMode
+    && !props.projectId
+    && !store.streaming
+    && (store.currentConversation?.messages.length ?? 0) === 0
+)
+
+const recentConversations = computed(() => {
+  const currentId = store.currentConversation?.id
+  return store.conversations
+    .filter((c) => c.id !== currentId && c.message_count > 0)
+    .slice(0, 5)
+})
+
+function startVoiceMode() {
+  window.dispatchEvent(new CustomEvent('voice:ptt-toggle'))
+}
+
 // ── Send ──────────────────────────────────────────────────────────────────────
 const inputBarRef = ref<InstanceType<typeof ChatInputBar> | null>(null)

@@ -271,8 +279,7 @@ async function handleSaveAsNote(messageId: number) {
 }

 // ── Lifecycle ─────────────────────────────────────────────────────────────────
-onMounted(async () => {
-  if (showScopeChip.value) await loadProjects()
+onMounted(() => {
  if (props.autoFocus) nextTick(() => inputBarRef.value?.focus())
 })

@@ -306,13 +313,13 @@ function hasEarlierBriefingSlot(index: number): boolean {
  return false
 }

-defineExpose({ focus, prefill, send })
+defineExpose({ focus, prefill, send, contextCount, sidebarOpen, toggleContextSidebar, hasContextData })
 </script>

 <template>
  <!-- ═══════════════════════════════ FULL VARIANT ══════════════════════════════ -->
  <template v-if="variant === 'full'">
-    <div class="chat-body" :class="{ 'chat-body--has-sidebar': hasContextSidebar }">
+    <div class="chat-full">
      <!-- Message list -->
      <div ref="messagesEl" class="messages-container">
        <div class="messages-inner">
@@ -350,8 +357,27 @@ defineExpose({ focus, prefill, send })
              </button>
            </div>
          </template>
+          <div v-if="showEmptyState" class="chat-empty-state">
+            <h2 class="empty-greeting">What's on your mind?</h2>
+            <div v-if="recentConversations.length" class="empty-recent">
+              <div class="empty-section-label">Jump back in</div>
+              <router-link
+                v-for="conv in recentConversations"
+                :key="conv.id"
+                :to="`/chat/${conv.id}`"
+                class="empty-recent-item"
+              >
+                <span class="empty-recent-title">{{ conv.title || 'Untitled conversation' }}</span>
+                <span class="empty-recent-count">{{ conv.message_count }} msg</span>
+              </router-link>
+            </div>
+            <button class="empty-voice-btn" @click="startVoiceMode" title="Start in voice mode">
+              <span class="voice-icon">🎤</span>
+              <span>Start in voice mode</span>
+            </button>
+          </div>
          <p
-            v-if="!store.currentConversation?.messages.length && !store.streaming"
+            v-else-if="!store.currentConversation?.messages.length && !store.streaming"
            class="empty-msg"
          >Start a conversation.</p>
        </div>
@@ -360,114 +386,71 @@ defineExpose({ focus, prefill, send })
      <!-- Context sidebar (full, non-briefing, non-workspace) -->
      <aside v-if="hasContextSidebar" class="context-sidebar">
        <template v-if="autoInjectedNotes.length">
-          <div class="context-sidebar-header">Auto-included</div>
-          <div v-for="note in autoInjectedNotes" :key="note.id" class="context-note context-note-auto">
-            <router-link :to="`/notes/${note.id}`" class="context-note-name">{{ note.title }}</router-link>
-            <span
-              v-if="note.score != null"
-              class="context-note-score"
-              :class="{ 'score-high': note.score >= 0.75, 'score-medium': note.score >= 0.60 && note.score < 0.75, 'score-low': note.score < 0.60 }"
-            >{{ Math.round(note.score * 100) }}%</span>
-            <button class="context-note-remove" @click="excludeAutoNote(note.id)">&times;</button>
+          <button
+            class="context-sidebar-header"
+            :class="{ collapsed: collapsedSections.has('auto') }"
+            @click="toggleSection('auto')"
+          >
+            <svg class="chev" width="10" height="10" viewBox="0 0 10 10" fill="currentColor"><path d="M2 3l3 4 3-4z"/></svg>
+            <span>Auto-included</span>
+            <span class="section-count">{{ autoInjectedNotes.length }}</span>
+          </button>
+          <div v-show="!collapsedSections.has('auto')">
+            <div v-for="note in autoInjectedNotes" :key="note.id" class="context-note context-note-auto">
+              <span class="auto-pill" title="Auto-included by relevance">AUTO</span>
+              <router-link :to="`/notes/${note.id}`" class="context-note-name">{{ note.title }}</router-link>
+              <span
+                v-if="note.score != null"
+                class="context-note-score"
+                :class="{ 'score-high': note.score >= 0.75, 'score-medium': note.score >= 0.60 && note.score < 0.75, 'score-low': note.score < 0.60 }"
+              >{{ Math.round(note.score * 100) }}%</span>
+              <button class="context-note-remove" @click="excludeAutoNote(note.id)" title="Remove from context">&times;</button>
+            </div>
          </div>
        </template>
        <template v-if="suggestedNotes.length">
-          <div class="context-sidebar-header" :class="{ 'context-sidebar-header-gap': autoInjectedNotes.length }">Suggested</div>
-          <div v-for="note in suggestedNotes" :key="note.id" class="context-note context-note-suggested">
-            <router-link :to="`/notes/${note.id}`" class="context-note-name">{{ note.title }}</router-link>
-            <span
-              v-if="note.score != null"
-              class="context-note-score"
-              :class="{ 'score-high': note.score >= 0.75, 'score-medium': note.score >= 0.60 && note.score < 0.75, 'score-low': note.score < 0.60 }"
-            >{{ Math.round(note.score * 100) }}%</span>
-            <button class="context-note-add" @click="includeNote(note)" title="Add to context">+</button>
+          <button
+            class="context-sidebar-header"
+            :class="{ collapsed: collapsedSections.has('suggested'), 'context-sidebar-header-gap': autoInjectedNotes.length }"
+            @click="toggleSection('suggested')"
+          >
+            <svg class="chev" width="10" height="10" viewBox="0 0 10 10" fill="currentColor"><path d="M2 3l3 4 3-4z"/></svg>
+            <span>Suggested</span>
+            <span class="section-count">{{ suggestedNotes.length }}</span>
+          </button>
+          <div v-show="!collapsedSections.has('suggested')">
+            <div v-for="note in suggestedNotes" :key="note.id" class="context-note context-note-suggested">
+              <router-link :to="`/notes/${note.id}`" class="context-note-name">{{ note.title }}</router-link>
+              <span
+                v-if="note.score != null"
+                class="context-note-score"
+                :class="{ 'score-high': note.score >= 0.75, 'score-medium': note.score >= 0.60 && note.score < 0.75, 'score-low': note.score < 0.60 }"
+              >{{ Math.round(note.score * 100) }}%</span>
+              <button class="context-note-add" @click="includeNote(note)" title="Add to context">+</button>
+            </div>
          </div>
        </template>
        <template v-if="includedNotes.length">
-          <div class="context-sidebar-header" :class="{ 'context-sidebar-header-gap': autoInjectedNotes.length || suggestedNotes.length }">In Context</div>
-          <div v-for="note in includedNotes" :key="note.id" class="context-note context-note-included">
-            <router-link :to="`/notes/${note.id}`" class="context-note-name">{{ note.title }}</router-link>
-            <button class="context-note-remove" @click="removeIncludedNote(note.id)">&times;</button>
+          <button
+            class="context-sidebar-header"
+            :class="{ collapsed: collapsedSections.has('included'), 'context-sidebar-header-gap': autoInjectedNotes.length || suggestedNotes.length }"
+            @click="toggleSection('included')"
+          >
+            <svg class="chev" width="10" height="10" viewBox="0 0 10 10" fill="currentColor"><path d="M2 3l3 4 3-4z"/></svg>
+            <span>In Context</span>
+            <span class="section-count">{{ includedNotes.length }}</span>
+          </button>
+          <div v-show="!collapsedSections.has('included')">
+            <div v-for="note in includedNotes" :key="note.id" class="context-note context-note-included">
+              <router-link :to="`/notes/${note.id}`" class="context-note-name">{{ note.title }}</router-link>
+              <button class="context-note-remove" @click="removeIncludedNote(note.id)" title="Remove from context">&times;</button>
+            </div>
          </div>
        </template>
      </aside>
-    </div>
-
-    <!-- Input area (hidden when readOnly) -->
-    <div v-if="!readOnly" class="input-wrapper">
-      <!-- Scope chip -->
-      <div v-if="showScopeChip" class="scope-chip-row">
-        <div class="scope-chip-wrapper">
-          <button
-            class="scope-chip"
-            :class="{ pulse: scopePulse }"
-            @click="scopeDropdownOpen = !scopeDropdownOpen"
-            title="Change RAG scope"
-          >
-            <span class="scope-dot">⊙</span> {{ scopeLabel }}
-          </button>
-          <div v-if="scopeDropdownOpen" class="scope-dropdown">
-            <button class="scope-option" :class="{ active: store.ragProjectId === null }" @click="onScopeSelect(null)">Orphan notes only</button>
-            <button
-              v-for="p in projects"
-              :key="p.id"
-              class="scope-option"
-              :class="{ active: store.ragProjectId === p.id }"
-              @click="onScopeSelect(p.id)"
-            >{{ p.title }}</button>
-            <button class="scope-option" :class="{ active: store.ragProjectId === -1 }" @click="onScopeSelect(-1)">All notes</button>
-          </div>
-        </div>
-      </div>
-
-      <!-- Listen / volume controls -->
-      <div v-if="voiceTtsEnabled" class="voice-controls">
-        <div class="volume-wrapper">
-          <button
-            class="btn-voice"
-            :class="{ 'btn-voice--active': listenMode, 'btn-voice--busy': tts.speaking.value }"
-            @click="toggleListen"
-            :title="listenMode ? 'Stop auto-read' : 'Read responses aloud'"
-            aria-label="Toggle listen mode"
-          >
-            <svg v-if="!tts.speaking.value" width="17" height="17" viewBox="0 0 24 24" fill="currentColor">
-              <path d="M3 9v6h4l5 5V4L7 9H3zm13.5 3c0-1.77-1.02-3.29-2.5-4.03v8.05c1.48-.73 2.5-2.25 2.5-4.02zM14 3.23v2.06c2.89.86 5 3.54 5 6.71s-2.11 5.85-5 6.71v2.06c4.01-.91 7-4.49 7-8.77s-2.99-7.86-7-8.77z"/>
-            </svg>
-            <svg v-else width="17" height="17" viewBox="0 0 24 24" fill="currentColor">
-              <path d="M18 12c0-1.77-1.02-3.29-2.5-4.03v2.21l2.45 2.45c.03-.2.05-.41.05-.63zm2.5 0c0 .94-.2 1.82-.54 2.64l1.51 1.51C21.8 14.82 22 13.43 22 12c0-4.28-2.99-7.86-7-8.77v2.06c2.89.86 5 3.54 5 6.71zM4.27 3L3 4.27 7.73 9H3v6h4l5 5v-6.73l4.25 4.25c-.67.52-1.42.93-2.25 1.18v2.06c1.38-.31 2.63-.95 3.69-1.81L19.73 21 21 19.73l-9-9L4.27 3zM12 4L9.91 6.09 12 8.18V4z"/>
-            </svg>
-          </button>
-          <button
-            class="btn-voice btn-volume-icon"
-            @click="showVolumeSlider = !showVolumeSlider"
-            :title="`Volume: ${Math.round(audio.volume.value * 100)}%`"
-            aria-label="Volume"
-          >
-            <svg width="14" height="14" viewBox="0 0 24 24" fill="currentColor">
-              <path d="M3 9v6h4l5 5V4L7 9H3zm13.5 3c0-1.77-1.02-3.29-2.5-4.03v8.05c1.48-.73 2.5-2.25 2.5-4.02z"/>
-            </svg>
-          </button>
-          <div v-if="showVolumeSlider" class="volume-popup">
-            <input
-              type="range" min="0" max="1" step="0.05"
-              :value="audio.volume.value"
-              @input="setVoiceVolume(+($event.target as HTMLInputElement).value)"
-              class="volume-range"
-              aria-label="Volume"
-            />
-            <span class="volume-pct">{{ Math.round(audio.volume.value * 100) }}%</span>
-          </div>
-        </div>
-        <button
-          v-if="tts.speaking.value"
-          class="btn-voice"
-          @click="tts.stop()"
-          title="Stop playback"
-        >
-          <svg width="13" height="13" viewBox="0 0 14 14" fill="currentColor"><rect width="14" height="14" rx="2"/></svg>
-        </button>
-      </div>

+      <!-- Input area (hidden when readOnly) -->
+      <div v-if="!readOnly" class="input-wrapper">
      <!-- Unified input bar -->
      <ChatInputBar
        ref="inputBarRef"
@@ -476,6 +459,7 @@ defineExpose({ focus, prefill, send })
        @submit="onSubmit"
        @abort="store.cancelGeneration()"
      />
+      </div>
    </div>
  </template>

@@ -528,18 +512,31 @@ defineExpose({ focus, prefill, send })
 </template>

 <style scoped>
-/* ── Full variant layout ── */
-.chat-body {
+/* ── Full variant layout ──
+ * Single grid owns the reading column + context sidebar + input bar so
+ * messages and input bar share one centered reading track while the
+ * context sidebar spans the full height from header to bottom of view.
+ * Reading column is always centered (3-column grid: 1fr | content | 1fr).
+ * The context sidebar overlays the right gutter so it never shifts layout.
+ */
+.chat-full {
+  position: relative;
  flex: 1;
-  display: flex;
  min-height: 0;
+  display: grid;
+  grid-template-columns:
+    1fr
+    minmax(0, var(--chat-reading-width))
+    1fr;
+  grid-template-rows: minmax(0, 1fr) auto;
  overflow: hidden;
 }

 .messages-container {
-  flex: 1;
+  grid-column: 2;
+  grid-row: 1;
  overflow-y: auto;
-  padding: 1rem;
+  padding: 1rem 1rem 0.75rem;
  mask-image: linear-gradient(to bottom, transparent, black 20px, black calc(100% - 20px), transparent);
  -webkit-mask-image: linear-gradient(to bottom, transparent, black 20px, black calc(100% - 20px), transparent);
 }
@@ -558,31 +555,79 @@ defineExpose({ focus, prefill, send })
  opacity: 0.35;
 }

-/* Context sidebar */
+/* Context sidebar — overlays right gutter, never shifts reading column */
 .context-sidebar {
-  width: 200px;
-  min-width: 160px;
-  max-width: 220px;
+  position: absolute;
+  top: 0;
+  right: 0;
+  bottom: 0;
+  width: var(--chat-context-sidebar-width);
  border-left: 1px solid var(--color-border);
  padding: 0.75rem 0.5rem;
  overflow-y: auto;
  background: var(--color-bg-secondary);
  font-size: 0.78rem;
+  z-index: 2;
 }
 .context-sidebar-header {
+  display: flex;
+  align-items: center;
+  gap: 0.35rem;
+  width: 100%;
+  background: none;
+  border: none;
+  padding: 0.15rem 0.1rem;
+  font-family: inherit;
  font-size: 0.65rem;
  font-weight: 700;
  text-transform: uppercase;
  letter-spacing: 0.05em;
  color: var(--color-text-muted);
  margin-bottom: 0.35rem;
+  cursor: pointer;
+  text-align: left;
+}
+.context-sidebar-header:hover { color: var(--color-text); }
+.context-sidebar-header .chev {
+  transition: transform 0.15s ease;
+}
+.context-sidebar-header.collapsed .chev {
+  transform: rotate(-90deg);
+}
+.context-sidebar-header .section-count {
+  margin-left: auto;
+  color: var(--color-text-muted);
+  font-weight: 600;
+  background: var(--color-bg);
+  padding: 0.05rem 0.3rem;
+  border-radius: 8px;
 }
 .context-sidebar-header-gap { margin-top: 0.75rem; }
 .context-note {
  display: flex;
  align-items: center;
-  gap: 0.2rem;
+  gap: 0.25rem;
  margin-bottom: 0.25rem;
+  padding: 0.2rem 0.35rem;
+  border-radius: var(--radius-sm);
+}
+.context-note-auto {
+  background: color-mix(in srgb, var(--color-text-muted) 8%, transparent);
+  opacity: 0.85;
+}
+.context-note-included {
+  background: color-mix(in srgb, var(--color-primary) 10%, transparent);
+  border-left: 2px solid var(--color-primary);
+}
+.auto-pill {
+  font-size: 0.55rem;
+  font-weight: 700;
+  letter-spacing: 0.05em;
+  padding: 0.05rem 0.3rem;
+  border-radius: 4px;
+  background: var(--color-bg);
+  color: var(--color-text-muted);
+  flex-shrink: 0;
 }
 .context-note-name {
  flex: 1;
@@ -616,8 +661,10 @@ defineExpose({ focus, prefill, send })
 .context-note-remove:hover { color: #ef4444; }
 .context-note-add:hover { color: var(--color-primary); }

-/* Input wrapper */
+/* Input wrapper — lives in the same reading column as messages */
 .input-wrapper {
+  grid-column: 2;
+  grid-row: 2;
  border-top: 1px solid var(--color-border);
  padding: 0.5rem 1rem 0.75rem;
  display: flex;
@@ -625,94 +672,6 @@ defineExpose({ focus, prefill, send })
  gap: 0.4rem;
 }

-/* Scope chip */
-.scope-chip-row { display: flex; }
-.scope-chip-wrapper { position: relative; }
-.scope-chip {
-  display: flex;
-  align-items: center;
-  gap: 0.3rem;
-  background: none;
-  border: 1px solid var(--color-border);
-  border-radius: 20px;
-  padding: 0.2rem 0.65rem;
-  font-size: 0.75rem;
-  color: var(--color-text-muted);
-  cursor: pointer;
-  transition: border-color 0.15s, color 0.15s;
-}
-.scope-chip:hover { border-color: var(--color-primary); color: var(--color-primary); }
-.scope-chip.pulse { animation: pulse-chip 0.6s ease; }
-@keyframes pulse-chip {
-  0%, 100% { border-color: var(--color-border); }
-  50% { border-color: var(--color-primary); color: var(--color-primary); box-shadow: 0 0 6px rgba(99,102,241,0.4); }
-}
-.scope-dot { font-size: 0.6rem; }
-.scope-dropdown {
-  position: absolute;
-  top: calc(100% + 4px);
-  left: 0;
-  min-width: 180px;
-  background: var(--color-bg-card);
-  border: 1px solid var(--color-border);
-  border-radius: var(--radius-md);
-  box-shadow: 0 4px 16px var(--color-shadow);
-  z-index: 20;
-  overflow: hidden;
-}
-.scope-option {
-  display: block;
-  width: 100%;
-  padding: 0.45rem 0.75rem;
-  text-align: left;
-  background: none;
-  border: none;
-  font-size: 0.85rem;
-  color: var(--color-text);
-  cursor: pointer;
-}
-.scope-option:hover { background: var(--color-bg-secondary); }
-.scope-option.active { color: var(--color-primary); font-weight: 600; }
-
-/* Voice controls */
-.voice-controls {
-  display: flex;
-  align-items: center;
-  gap: 0.2rem;
-}
-.volume-wrapper { position: relative; display: flex; align-items: center; gap: 0.1rem; }
-.btn-voice {
-  background: none;
-  border: none;
-  cursor: pointer;
-  color: var(--color-text-muted);
-  opacity: 0.65;
-  padding: 0.2rem;
-  display: flex;
-  align-items: center;
-  border-radius: var(--radius-sm);
-  transition: opacity 0.15s, color 0.15s;
-}
-.btn-voice:hover { opacity: 1; }
-.btn-voice--active { opacity: 1; color: var(--color-primary); }
-.btn-voice--busy { opacity: 1; color: #f59e0b; }
-.volume-popup {
-  position: absolute;
-  bottom: calc(100% + 8px);
-  left: 0;
-  background: var(--color-bg-card);
-  border: 1px solid var(--color-border);
-  border-radius: var(--radius-md);
-  padding: 0.5rem 0.75rem;
-  display: flex;
-  align-items: center;
-  gap: 0.5rem;
-  z-index: 20;
-  box-shadow: 0 4px 16px var(--color-shadow);
-}
-.volume-range { width: 80px; }
-.volume-pct { font-size: 0.75rem; color: var(--color-text-muted); white-space: nowrap; }
-
 /* Queued messages */
 .queued-message { opacity: 0.45; }
 .queued-bubble {
@@ -754,6 +713,108 @@ defineExpose({ focus, prefill, send })
  color: var(--color-accent-warm, #d4a017);
 }

+.chat-empty-state {
+  display: flex;
+  flex-direction: column;
+  align-items: stretch;
+  gap: 1.5rem;
+  padding: 3rem 1rem 2rem;
+  max-width: 520px;
+  margin: 0 auto;
+}
+
+.empty-greeting {
+  font-family: 'Fraunces', Georgia, serif;
+  font-style: italic;
+  font-weight: 400;
+  font-size: 1.75rem;
+  color: var(--color-accent-warm, #d4a017);
+  text-align: center;
+  margin: 0;
+}
+
+.empty-section-label {
+  font-size: 0.7rem;
+  font-weight: 600;
+  text-transform: uppercase;
+  letter-spacing: 0.08em;
+  color: var(--color-text-muted);
+  margin-bottom: 0.5rem;
+  padding-left: 0.25rem;
+}
+
+.empty-recent {
+  display: flex;
+  flex-direction: column;
+}
+
+.empty-recent-item {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 0.75rem;
+  padding: 0.75rem 0.9rem;
+  border-radius: var(--radius-md, 10px);
+  border: 1px solid var(--color-border);
+  background: var(--color-bg-elevated, rgba(255, 255, 255, 0.02));
+  color: var(--color-text);
+  text-decoration: none;
+  font-size: 0.9rem;
+  transition: border-color 0.15s ease, background 0.15s ease, transform 0.15s ease;
+}
+
+.empty-recent-item + .empty-recent-item {
+  margin-top: 0.4rem;
+}
+
+.empty-recent-item:hover {
+  border-color: var(--color-primary);
+  background: rgba(99, 102, 241, 0.05);
+  transform: translateX(2px);
+}
+
+.empty-recent-title {
+  overflow: hidden;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+  flex: 1;
+  min-width: 0;
+}
+
+.empty-recent-count {
+  font-size: 0.7rem;
+  color: var(--color-text-muted);
+  font-variant-numeric: tabular-nums;
+  flex-shrink: 0;
+}
+
+.empty-voice-btn {
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  gap: 0.6rem;
+  padding: 0.75rem 1.2rem;
+  border-radius: 999px;
+  border: 1px solid var(--color-border);
+  background: transparent;
+  color: var(--color-text);
+  font-size: 0.9rem;
+  cursor: pointer;
+  align-self: center;
+  transition: border-color 0.15s ease, background 0.15s ease, color 0.15s ease;
+}
+
+.empty-voice-btn:hover {
+  border-color: var(--color-primary);
+  background: linear-gradient(135deg, #6366f1, #4f46e5);
+  color: #fff;
+  box-shadow: 0 4px 16px rgba(99, 102, 241, 0.35);
+}
+
+.empty-voice-btn .voice-icon {
+  font-size: 1.05rem;
+}
+
 /* ── Widget variant ── */
 .widget-response {
  margin-top: 0.75rem;
@@ -1,15 +1,54 @@
 import { ref, readonly } from 'vue'

 export interface SilenceDetectorOptions {
-  thresholdDb?: number       // default -40
-  silenceDurationMs?: number // default 1500
-  minRecordingMs?: number    // default 500
+  /**
+   * Absolute fallback silence threshold in dBFS, used during the first
+   * moments before any real speech has been observed. Once the session peak
+   * clears `dynamicArmDb` we switch to the dynamic threshold instead.
+   */
+  fallbackThresholdDb?: number // default -45
+  /** How many dB below the session peak counts as "silent" once armed. */
+  dropFromPeakDb?: number // default 15
+  /**
+   * Session peak must reach this level (dBFS) before dynamic thresholding
+   * kicks in. Until then the fallback threshold is used so we don't lock
+   * onto a noise-floor peak.
+   */
+  dynamicArmDb?: number // default -25
+  /** How long to wait at the start of recording before running silence checks. */
+  graceMs?: number // default 1500
+  silenceDurationMs?: number // default 2000
+  minRecordingMs?: number // default 500
 }

+/**
+ * Mic silence detector + live amplitude signal.
+ *
+ * Uses `getFloatTimeDomainData()` for honest linear RMS in [-1, 1] space,
+ * then derives dBFS for the silence threshold. The previous implementation
+ * ran RMS over `getByteFrequencyData` bytes — those bytes are already a
+ * dB-scaled quantity, so taking their RMS and re-log'ing it produced
+ * numbers that didn't line up with real dBFS and made any static threshold
+ * unpredictable.
+ *
+ * Silence threshold is dynamic: we track the session peak dBFS and treat
+ * "silent" as "current level is at least [dropFromPeakDb] below peak."
+ * This auto-calibrates to whatever mic and room the user is on. Until the
+ * peak climbs above [dynamicArmDb] we fall back to a conservative static
+ * threshold so a quiet room doesn't spin forever. A grace period at the
+ * start of the recording gives the user time to begin speaking before
+ * silence checks arm.
+ *
+ * `amplitude` is the raw linear RMS scaled ×5 and clamped to 1 so quiet
+ * speech still visibly moves UI that binds to it.
+ */
 export function useSilenceDetector(options: SilenceDetectorOptions = {}) {
  const {
-    thresholdDb = -40,
-    silenceDurationMs = 1500,
+    fallbackThresholdDb = -45,
+    dropFromPeakDb = 15,
+    dynamicArmDb = -25,
+    graceMs = 1500,
+    silenceDurationMs = 2000,
    minRecordingMs = 500,
  } = options

@@ -18,31 +57,43 @@ export function useSilenceDetector(options: SilenceDetectorOptions = {}) {
  let intervalId: ReturnType<typeof setInterval> | null = null
  let silenceMs = 0
  let startedAt = 0
+  let peakDb = -100

  function start(stream: MediaStream, onSilence: () => void): void {
    stop()
    audioCtx = new AudioContext()
    // Some browsers start AudioContext in "suspended" state — resume so
-    // getByteFrequencyData returns real values instead of all zeros.
+    // the analyser returns real values instead of all zeros.
    audioCtx.resume().catch(() => {})
    const source = audioCtx.createMediaStreamSource(stream)
    const analyser = audioCtx.createAnalyser()
-    analyser.fftSize = 256
+    analyser.fftSize = 1024
    source.connect(analyser)

-    const data = new Uint8Array(analyser.frequencyBinCount)
+    const samples = new Float32Array(analyser.fftSize)
    silenceMs = 0
    startedAt = Date.now()
+    peakDb = -100

    intervalId = setInterval(() => {
-      analyser.getByteFrequencyData(data)
-      const rms = Math.sqrt(data.reduce((s, v) => s + v * v, 0) / data.length) / 255
-      amplitude.value = rms
+      analyser.getFloatTimeDomainData(samples)
+      let sumSq = 0
+      for (let i = 0; i < samples.length; i++) sumSq += samples[i] * samples[i]
+      const rms = Math.sqrt(sumSq / samples.length)
+      amplitude.value = Math.min(rms * 5, 1)

-      const db = rms > 0 ? 20 * Math.log10(rms) : -100
-      if (db < thresholdDb) {
+      const db = rms > 1e-7 ? 20 * Math.log10(rms) : -100
+      if (db > peakDb) peakDb = db
+
+      const elapsed = Date.now() - startedAt
+      if (elapsed < graceMs) return
+
+      const threshold =
+        peakDb > dynamicArmDb ? peakDb - dropFromPeakDb : fallbackThresholdDb
+
+      if (db < threshold) {
        silenceMs += 100
-        if (silenceMs >= silenceDurationMs && Date.now() - startedAt >= minRecordingMs) {
+        if (silenceMs >= silenceDurationMs && elapsed >= minRecordingMs) {
          stop()
          onSilence()
        }
@@ -63,6 +114,7 @@ export function useSilenceDetector(options: SilenceDetectorOptions = {}) {
    }
    amplitude.value = 0
    silenceMs = 0
+    peakDb = -100
  }

  return { amplitude: readonly(amplitude), start, stop }
@@ -2,6 +2,7 @@
 import { onMounted, onUnmounted, ref, computed, watch, nextTick } from "vue";
 import { useRoute, useRouter } from "vue-router";
 import { useChatStore } from "@/stores/chat";
+import { apiGet } from "@/api/client";
 import ChatPanel from "@/components/ChatPanel.vue";

 const route = useRoute();
@@ -10,6 +11,49 @@ const store = useChatStore();

 const summarizing = ref(false);
 const sidebarOpen = ref(false);
+const convSearchQuery = ref("");
+const headerKebabOpen = ref(false);
+const sidebarKebabOpen = ref(false);
+
+// ── RAG scope chip ────────────────────────────────────────────────────────────
+const projects = ref<{ id: number; title: string }[]>([]);
+const scopeDropdownOpen = ref(false);
+const scopePulse = ref(false);
+
+const scopeLabel = computed(() => {
+  const id = store.ragProjectId;
+  if (id === -1) return "All notes";
+  if (id === null) return "Orphan notes";
+  return projects.value.find((p) => p.id === id)?.title ?? `Project ${id}`;
+});
+
+async function loadProjects() {
+  try {
+    const data = await apiGet<{ projects: { id: number; title: string }[] }>(
+      "/api/projects?status=active"
+    );
+    projects.value = data.projects ?? [];
+  } catch {
+    projects.value = [];
+  }
+}
+
+async function onScopeSelect(value: number | null) {
+  scopeDropdownOpen.value = false;
+  const id = store.currentConversation?.id;
+  if (!id) return;
+  await store.updateRagScope(id, value);
+  scopePulse.value = true;
+  setTimeout(() => { scopePulse.value = false; }, 600);
+}
+
+watch(
+  () => store.ragProjectId,
+  () => {
+    scopePulse.value = true;
+    setTimeout(() => { scopePulse.value = false; }, 600);
+  }
+);

 let prevConvId: number | null = null;

@@ -46,7 +90,12 @@ const groupedConversations = computed((): ConvGroup[] => {
  const buckets: Record<string, typeof store.conversations> = {};
  const order: string[] = [];

-  for (const conv of store.conversations) {
+  const q = convSearchQuery.value.trim().toLowerCase();
+  const filtered = q
+    ? store.conversations.filter((c) => (c.title || "").toLowerCase().includes(q))
+    : store.conversations;
+
+  for (const conv of filtered) {
    const d = new Date(conv.updated_at);
    let key: string;
    if (d >= startOfToday) {
@@ -69,6 +118,8 @@ const groupedConversations = computed((): ConvGroup[] => {

 onMounted(async () => {
  document.addEventListener("keydown", onGlobalKeydown);
+  document.addEventListener("mousedown", onDocumentMousedown);
+  loadProjects();
  await store.fetchConversations();
  if (convId.value) {
    if (store.currentConversation?.id !== convId.value) {
@@ -187,36 +238,35 @@ async function handleSummarize() {
  }
 }

-// ── Research modal ────────────────────────────────────────────────────────────
-const showResearchModal = ref(false);
-const researchTopic = ref("");
 const chatPanelRef = ref<InstanceType<typeof ChatPanel> | null>(null);

-function toggleResearchModal() {
-  showResearchModal.value = !showResearchModal.value;
-  if (showResearchModal.value) {
-    researchTopic.value = "";
-    nextTick(() => {
-      const el = document.querySelector(".research-topic-input") as HTMLInputElement;
-      el?.focus();
-    });
-  }
+const contextCount = computed(() => chatPanelRef.value?.contextCount ?? 0);
+const contextSidebarOpen = computed(() => chatPanelRef.value?.sidebarOpen ?? false);
+function toggleContextSidebar() {
+  chatPanelRef.value?.toggleContextSidebar();
 }

-function submitResearch() {
-  const topic = researchTopic.value.trim();
-  if (!topic) return;
-  showResearchModal.value = false;
-  researchTopic.value = "";
-  // Prefill sends via ChatPanel — user sees it in the input, then it auto-submits
-  chatPanelRef.value?.send(`Research: ${topic}`);
+// ── Kebab dismissal ───────────────────────────────────────────────────────────
+function onDocumentMousedown(e: MouseEvent) {
+  const target = e.target as HTMLElement | null;
+  if (headerKebabOpen.value && !target?.closest(".header-kebab-wrapper")) {
+    headerKebabOpen.value = false;
+  }
+  if (sidebarKebabOpen.value && !target?.closest(".sidebar-kebab-wrapper")) {
+    sidebarKebabOpen.value = false;
+  }
+  if (scopeDropdownOpen.value && !target?.closest(".scope-chip-wrapper")) {
+    scopeDropdownOpen.value = false;
+  }
 }

 // ── Keyboard ──────────────────────────────────────────────────────────────────
 function onGlobalKeydown(e: KeyboardEvent) {
  if (e.key !== "Escape") return;
-  if (showResearchModal.value) {
-    showResearchModal.value = false;
+  if (headerKebabOpen.value || sidebarKebabOpen.value || scopeDropdownOpen.value) {
+    headerKebabOpen.value = false;
+    sidebarKebabOpen.value = false;
+    scopeDropdownOpen.value = false;
    return;
  }
  if (sidebarOpen.value) {
@@ -228,6 +278,7 @@ function onGlobalKeydown(e: KeyboardEvent) {

 onUnmounted(() => {
  document.removeEventListener("keydown", onGlobalKeydown);
+  document.removeEventListener("mousedown", onDocumentMousedown);
  if (prevConvId) {
    const conv = store.conversations.find((c) => c.id === prevConvId);
    if (conv && conv.message_count === 0) {
@@ -246,13 +297,35 @@ onUnmounted(() => {
    ></div>
    <aside class="chat-sidebar" :class="{ open: sidebarOpen }">
      <div class="sidebar-top-bar">
-        <button class="btn-new-conv" @click="newConversation">+ New Chat</button>
-        <button
-          class="btn-select-mode"
-          :class="{ active: selectMode }"
-          @click="toggleSelectMode"
-          title="Select conversations"
-        >Select</button>
+        <button class="btn-new-conv btn-new-conv--full" @click="newConversation">+ New Chat</button>
+        <div class="sidebar-search-row">
+          <input
+            v-model="convSearchQuery"
+            class="conv-search-input"
+            type="search"
+            placeholder="Search conversations…"
+            aria-label="Search conversations"
+          />
+          <div class="sidebar-kebab-wrapper">
+            <button
+              class="btn-kebab"
+              :class="{ active: sidebarKebabOpen }"
+              @click="sidebarKebabOpen = !sidebarKebabOpen"
+              aria-label="List actions"
+              title="List actions"
+            >
+              <svg width="18" height="18" viewBox="0 0 24 24" fill="currentColor">
+                <circle cx="12" cy="5" r="1.75"/><circle cx="12" cy="12" r="1.75"/><circle cx="12" cy="19" r="1.75"/>
+              </svg>
+            </button>
+            <div v-if="sidebarKebabOpen" class="kebab-menu">
+              <button
+                class="kebab-item"
+                @click="sidebarKebabOpen = false; toggleSelectMode()"
+              >{{ selectMode ? "Exit select mode" : "Select conversations" }}</button>
+            </div>
+          </div>
+        </div>
      </div>

      <div v-if="selectMode" class="bulk-bar">
@@ -304,6 +377,10 @@ onUnmounted(() => {
        <p v-if="!store.conversations.length" class="empty-msg">
          No conversations yet
        </p>
+        <p
+          v-else-if="convSearchQuery && !groupedConversations.length"
+          class="empty-msg"
+        >No matches for “{{ convSearchQuery }}”</p>
      </div>
    </aside>

@@ -319,40 +396,68 @@ onUnmounted(() => {
          </button>
          <h2>{{ store.currentConversation.title || "New Chat" }}</h2>

-          <!-- Research modal trigger -->
-          <div class="research-wrapper">
+          <div class="scope-chip-wrapper">
            <button
-              class="btn-attach"
-              @click="toggleResearchModal"
-              :disabled="store.streaming || !store.chatReady"
-              title="Research a topic"
+              class="scope-chip"
+              :class="{ pulse: scopePulse }"
+              @click="scopeDropdownOpen = !scopeDropdownOpen"
+              title="Change RAG scope"
            >
-              <svg width="18" height="18" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
-                <circle cx="11" cy="11" r="8"/><line x1="21" y1="21" x2="16.65" y2="16.65"/>
-              </svg>
+              <span class="scope-dot">⊙</span> {{ scopeLabel }}
            </button>
-            <div v-if="showResearchModal" class="research-modal">
-              <div class="research-modal-header">Research topic</div>
-              <input
-                class="research-topic-input"
-                v-model="researchTopic"
-                placeholder="e.g. quantum computing"
-                @keydown.enter="submitResearch"
-                @keydown.escape="showResearchModal = false"
-              />
-              <div class="research-modal-actions">
-                <button class="btn-research-cancel" @click="showResearchModal = false">Cancel</button>
-                <button class="btn-research-go" @click="submitResearch" :disabled="!researchTopic.trim()">Go</button>
-              </div>
+            <div v-if="scopeDropdownOpen" class="scope-dropdown">
+              <button
+                class="scope-option"
+                :class="{ active: store.ragProjectId === null }"
+                @click="onScopeSelect(null)"
+              >Orphan notes only</button>
+              <button
+                v-for="p in projects"
+                :key="p.id"
+                class="scope-option"
+                :class="{ active: store.ragProjectId === p.id }"
+                @click="onScopeSelect(p.id)"
+              >{{ p.title }}</button>
+              <button
+                class="scope-option"
+                :class="{ active: store.ragProjectId === -1 }"
+                @click="onScopeSelect(-1)"
+              >All notes</button>
            </div>
          </div>

          <button
-            v-if="store.currentConversation.messages.length"
-            class="btn-summarize"
-            @click="handleSummarize"
-            :disabled="summarizing || store.streaming"
-          >{{ summarizing ? "Summarizing..." : "Summarize as Note" }}</button>
+            v-if="contextCount > 0"
+            class="btn-context-toggle"
+            :class="{ active: contextSidebarOpen }"
+            @click="toggleContextSidebar"
+            :title="contextSidebarOpen ? 'Hide context panel' : 'Show context panel'"
+          >
+            <span class="context-icon">📎</span>
+            <span class="context-label">Context</span>
+            <span class="context-count-badge">{{ contextCount }}</span>
+          </button>
+
+          <div class="header-kebab-wrapper">
+            <button
+              class="btn-kebab"
+              :class="{ active: headerKebabOpen }"
+              @click="headerKebabOpen = !headerKebabOpen"
+              aria-label="Conversation actions"
+              title="Conversation actions"
+            >
+              <svg width="18" height="18" viewBox="0 0 24 24" fill="currentColor">
+                <circle cx="12" cy="5" r="1.75"/><circle cx="12" cy="12" r="1.75"/><circle cx="12" cy="19" r="1.75"/>
+              </svg>
+            </button>
+            <div v-if="headerKebabOpen" class="kebab-menu kebab-menu--right">
+              <button
+                class="kebab-item"
+                :disabled="!store.currentConversation.messages.length || summarizing || store.streaming"
+                @click="headerKebabOpen = false; handleSummarize()"
+              >{{ summarizing ? "Summarizing…" : "Summarize as Note" }}</button>
+            </div>
+          </div>
        </div>

        <ChatPanel
@@ -397,12 +502,12 @@ onUnmounted(() => {

 .sidebar-top-bar {
  display: flex;
+  flex-direction: column;
  gap: 0.5rem;
  padding: 0.75rem;
 }

 .btn-new-conv {
-  flex: 1;
  padding: 0.5rem;
  background: var(--color-primary);
  color: #fff;
@@ -412,23 +517,31 @@ onUnmounted(() => {
  font-size: 0.9rem;
  transition: box-shadow 0.15s, opacity 0.15s;
 }
+.btn-new-conv--full { width: 100%; }
 .btn-new-conv:hover {
  opacity: 0.9;
  box-shadow: 0 0 14px rgba(129, 140, 248, 0.35);
 }

-.btn-select-mode {
-  background: none;
+.sidebar-search-row {
+  display: flex;
+  align-items: center;
+  gap: 0.35rem;
+}
+.conv-search-input {
+  flex: 1;
+  min-width: 0;
+  padding: 0.4rem 0.6rem;
  border: 1px solid var(--color-border);
  border-radius: var(--radius-sm);
-  color: var(--color-text-muted);
-  font-size: 0.82rem;
-  padding: 0.4rem 0.6rem;
-  cursor: pointer;
-  white-space: nowrap;
+  background: var(--color-bg);
+  color: var(--color-text);
+  font-size: 0.85rem;
+  outline: none;
+  font-family: inherit;
 }
-.btn-select-mode:hover,
-.btn-select-mode.active { border-color: var(--color-primary); color: var(--color-primary); }
+.conv-search-input:focus { border-color: var(--color-primary); }
+.conv-search-input::placeholder { color: var(--color-text-muted); }

 .bulk-bar {
  display: flex;
@@ -522,12 +635,12 @@ onUnmounted(() => {
  overflow: hidden;
 }

-/* ChatPanel fills the remaining space in chat-main */
+/* ChatPanel fills the remaining space in chat-main. Layout (grid vs
+ * flex) is owned by ChatPanel's own .chat-full root — don't force a
+ * display here or it overrides the grid. */
 .chat-panel-fill {
  flex: 1;
  min-height: 0;
-  display: flex;
-  flex-direction: column;
 }

 .chat-header {
@@ -548,87 +661,138 @@ onUnmounted(() => {
  min-width: 0;
 }

-.btn-summarize {
-  padding: 0.3rem 0.75rem;
-  background: var(--color-bg-secondary);
-  color: var(--color-text);
-  border: 1px solid var(--color-border);
-  border-radius: var(--radius-sm);
-  cursor: pointer;
-  font-size: 0.85rem;
-  white-space: nowrap;
-}
-.btn-summarize:hover:not(:disabled) {
-  background: var(--color-primary); color: #fff; border-color: var(--color-primary);
-}
-.btn-summarize:disabled { opacity: 0.6; cursor: default; }
-
-.btn-attach {
+/* Kebab button + dropdown menu — shared by header and sidebar */
+.btn-kebab {
  background: none;
  border: none;
  cursor: pointer;
  color: var(--color-text-muted);
-  opacity: 0.6;
-  padding: 0.2rem;
+  padding: 0.25rem;
  display: flex;
  align-items: center;
+  justify-content: center;
+  border-radius: var(--radius-sm);
+  flex-shrink: 0;
 }
-.btn-attach:hover { opacity: 1; }
-.btn-attach:disabled { opacity: 0.25; cursor: default; }
+.btn-kebab:hover { color: var(--color-text); background: var(--color-bg-secondary); }
+.btn-kebab.active { color: var(--color-primary); }

-.research-wrapper { position: relative; }
-.research-modal {
+.header-kebab-wrapper,
+.sidebar-kebab-wrapper {
+  position: relative;
+}
+
+/* RAG scope chip (header) */
+.scope-chip-wrapper { position: relative; flex-shrink: 0; }
+.scope-chip {
+  display: flex;
+  align-items: center;
+  gap: 0.3rem;
+  background: none;
+  border: 1px solid var(--color-border);
+  border-radius: 20px;
+  padding: 0.2rem 0.65rem;
+  font-size: 0.75rem;
+  color: var(--color-text-muted);
+  cursor: pointer;
+  transition: border-color 0.15s, color 0.15s;
+}
+.scope-chip:hover { border-color: var(--color-primary); color: var(--color-primary); }
+.scope-chip.pulse { animation: pulse-chip 0.6s ease; }
+@keyframes pulse-chip {
+  0%, 100% { border-color: var(--color-border); }
+  50% {
+    border-color: var(--color-primary);
+    color: var(--color-primary);
+    box-shadow: 0 0 6px rgba(99, 102, 241, 0.4);
+  }
+}
+.scope-dot { font-size: 0.6rem; }
+.scope-dropdown {
  position: absolute;
-  top: calc(100% + 8px);
+  top: calc(100% + 4px);
  right: 0;
+  min-width: 200px;
+  background: var(--color-bg-card);
+  border: 1px solid var(--color-border);
+  border-radius: var(--radius-md);
+  box-shadow: 0 4px 16px var(--color-shadow);
+  z-index: 20;
+  overflow: hidden;
+}
+.scope-option {
+  display: block;
+  width: 100%;
+  padding: 0.45rem 0.75rem;
+  text-align: left;
+  background: none;
+  border: none;
+  font-size: 0.85rem;
+  color: var(--color-text);
+  cursor: pointer;
+}
+.scope-option:hover { background: var(--color-bg-secondary); }
+.scope-option.active { color: var(--color-primary); font-weight: 600; }
+
+/* Context toggle button (header) */
+.btn-context-toggle {
+  display: flex;
+  align-items: center;
+  gap: 0.3rem;
+  background: none;
+  border: 1px solid var(--color-border);
+  border-radius: 20px;
+  padding: 0.2rem 0.55rem;
+  font-size: 0.75rem;
+  color: var(--color-text-muted);
+  cursor: pointer;
+  flex-shrink: 0;
+  transition: border-color 0.15s, color 0.15s, background 0.15s;
+  font-family: inherit;
+}
+.btn-context-toggle:hover { border-color: var(--color-primary); color: var(--color-primary); }
+.btn-context-toggle.active {
+  background: color-mix(in srgb, var(--color-primary) 10%, transparent);
+  border-color: var(--color-primary);
+  color: var(--color-primary);
+}
+.context-icon { font-size: 0.85rem; line-height: 1; }
+.context-count-badge {
+  font-size: 0.65rem;
+  font-weight: 700;
+  background: var(--color-bg);
+  padding: 0.05rem 0.35rem;
+  border-radius: 8px;
+  color: var(--color-text);
+}
+
+.kebab-menu {
+  position: absolute;
+  top: calc(100% + 4px);
+  left: 0;
+  min-width: 180px;
  background: var(--color-bg-card);
  border: 1px solid var(--color-border);
  border-radius: var(--radius-md);
  box-shadow: 0 4px 20px var(--color-shadow);
-  padding: 0.75rem;
-  min-width: 280px;
+  padding: 0.25rem;
  z-index: 20;
 }
-.research-modal-header {
-  font-size: 0.8rem;
-  font-weight: 600;
-  color: var(--color-text-muted);
-  margin-bottom: 0.5rem;
-  text-transform: uppercase;
-  letter-spacing: 0.05em;
-}
-.research-topic-input {
+.kebab-menu--right { left: auto; right: 0; }
+.kebab-item {
+  display: block;
  width: 100%;
-  padding: 0.45rem 0.65rem;
-  border: 1px solid var(--color-border);
-  border-radius: var(--radius-sm);
-  background: var(--color-bg);
+  text-align: left;
+  background: none;
+  border: none;
  color: var(--color-text);
-  font-size: 0.9rem;
-  outline: none;
-  font-family: inherit;
-  box-sizing: border-box;
+  font-size: 0.85rem;
+  padding: 0.45rem 0.65rem;
+  border-radius: var(--radius-sm);
+  cursor: pointer;
 }
-.research-topic-input:focus { border-color: var(--color-primary); }
-.research-modal-actions {
-  display: flex;
-  gap: 0.5rem;
-  margin-top: 0.5rem;
-  justify-content: flex-end;
-}
-.btn-research-cancel {
-  background: none; border: 1px solid var(--color-border);
-  border-radius: var(--radius-sm); padding: 0.3rem 0.65rem;
-  font-size: 0.85rem; cursor: pointer; color: var(--color-text-muted);
-}
-.btn-research-cancel:hover { border-color: var(--color-text-muted); }
-.btn-research-go {
-  background: var(--color-primary); color: #fff; border: none;
-  border-radius: var(--radius-sm); padding: 0.3rem 0.75rem;
-  font-size: 0.85rem; cursor: pointer; font-weight: 600;
-}
-.btn-research-go:disabled { opacity: 0.4; cursor: default; }
-.btn-research-go:not(:disabled):hover { opacity: 0.9; }
+.kebab-item:hover:not(:disabled) { background: var(--color-bg-secondary); color: var(--color-primary); }
+.kebab-item:disabled { opacity: 0.4; cursor: default; }

 .no-conversation {
  flex: 1;
@@ -1,6 +1,6 @@
 from datetime import datetime, timezone

-from sqlalchemy import ARRAY, DateTime, ForeignKey, Index, Integer, Text, UniqueConstraint
+from sqlalchemy import ARRAY, BigInteger, DateTime, ForeignKey, Index, Integer, Text, UniqueConstraint
 from sqlalchemy.orm import Mapped, mapped_column, relationship

 from fabledassistant.models import Base
@@ -46,6 +46,17 @@ class RssItem(Base):
    published_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), nullable=True)
    # Truncated to 2000 chars to keep DB size reasonable
    content: Mapped[str] = mapped_column(Text, default="")
+    # Full trafilatura-extracted article body, populated lazily on first
+    # discuss-click / enrichment pass. Nullable — most items never get this
+    # cached. Expires naturally with the item (90-day retention).
+    content_full: Mapped[str | None] = mapped_column(Text, nullable=True)
+    # Map-reduced conversation-ready context derived from content_full. See
+    # services/article_context.py — populated on first discuss click so
+    # repeat clicks skip both the fetch and the LLM map step.
+    context_prepared: Mapped[str | None] = mapped_column(Text, nullable=True)
+    content_fetched_at: Mapped[datetime | None] = mapped_column(
+        DateTime(timezone=True), nullable=True
+    )
    fetched_at: Mapped[datetime] = mapped_column(
        DateTime(timezone=True), default=lambda: datetime.now(timezone.utc)
    )
@@ -55,6 +66,13 @@ class RssItem(Base):
    classified_at: Mapped[datetime | None] = mapped_column(
        DateTime(timezone=True), nullable=True
    )
+    # Note persisting the first-click discussion summary. Set by the article
+    # discussion pipeline once the seeded chat completes its first assistant
+    # reply; links back into RAG so re-discussing the same article lands the
+    # prior summary in context.
+    discussion_note_id: Mapped[int | None] = mapped_column(
+        BigInteger, ForeignKey("notes.id", ondelete="SET NULL"), nullable=True
+    )

    feed: Mapped["RssFeed"] = relationship(back_populates="items")

@@ -532,25 +532,21 @@ async def discuss_article(item_id: int):
    if get_buffer(conv_id) is not None:
        return jsonify({"error": "Generation already in progress"}), 409

-    from fabledassistant.services.rss import _fetch_full_article
-    article_content = await _fetch_full_article(item.url) or item.content or ""
+    # Shared helper handles the three-layer cache (context_prepared →
+    # content_full → fresh fetch), writes the synthetic read_article tool
+    # exchange and the conversational seed user prompt into the conversation.
+    # The /news from-article route calls the same helper so behavior stays
+    # byte-identical across entry points.
+    from fabledassistant.services.article_context import (
+        EmptyArticleError,
+        seed_article_discussion,
+    )

-    # Store synthetic assistant message with read_article tool result
-    synthetic_tool_calls = [{
-        "function": "read_article",
-        "arguments": {"url": item.url},
-        "result": {
-            "success": True,
-            "type": "article_content",
-            "url": item.url,
-            "content": article_content,
-            "truncated": False,
-        },
-    }]
-    await add_message(conv_id, "assistant", "", status="complete", tool_calls=synthetic_tool_calls)
-
-    # Store user message
-    await add_message(conv_id, "user", "Please summarize and discuss this article.")
+    model = await get_setting(uid, "default_model", "") or ""
+    try:
+        discuss_prompt = await seed_article_discussion(conv_id, item, model)
+    except EmptyArticleError as e:
+        return jsonify({"error": str(e)}), 422

    # Reload conversation with fresh messages to build history
    conv = await get_conversation(uid, conv_id)
@@ -572,15 +568,13 @@ async def discuss_article(item_id: int):
        else:
            history.append(msg_dict)

-    model = await get_setting(uid, "default_model", "") or ""
-
    assistant_msg = await add_message(conv_id, "assistant", "", status="generating")
    buf = create_buffer(conv_id, assistant_msg.id)

    asyncio.create_task(run_generation(
        buf, history, model,
        uid, conv_id, conv.title or "",
-        "Please summarize and discuss this article.",
+        discuss_prompt,
    ))

    return jsonify({"assistant_message_id": assistant_msg.id, "status": "generating"}), 202
@@ -510,47 +510,78 @@ async def delete_model_route():
@chat_bp.route("/from-article/<int:item_id>", methods=["POST"])
@login_required
 async def create_conversation_from_article(item_id: int):
-    """Create a chat conversation seeded with an RSS article's content."""
+    """Create a chat conversation seeded for article discussion and auto-run.
+
+    Mirrors the briefing ``discuss_article`` route: creates a fresh
+    conversation, stages the shared synthetic read_article exchange + seed
+    prompt, then kicks off generation so the client lands on an in-flight
+    stream. The Flutter and web chat screens reconnect to the running buffer
+    on mount.
+    """
    from sqlalchemy import select as _select
    from fabledassistant.models import async_session as _async_session
-    from fabledassistant.models.conversation import Message
    from fabledassistant.models.rss_feed import RssItem, RssFeed
+    from fabledassistant.services.article_context import (
+        EmptyArticleError,
+        seed_article_discussion,
+    )

    uid = get_current_user_id()

    async with _async_session() as session:
        result = await session.execute(
-            _select(RssItem, RssFeed.title.label("feed_title"))
+            _select(RssItem)
            .join(RssFeed, RssItem.feed_id == RssFeed.id)
            .where(RssItem.id == item_id, RssFeed.user_id == uid)
        )
-        row = result.first()
+        item = result.scalars().first()

-    if row is None:
+    if item is None:
        return jsonify({"error": "Article not found"}), 404

-    item, feed_title = row
-
    conv_title = (item.title or "Article discussion")[:80]
    conv = await create_conversation(uid, title=conv_title, conversation_type="chat")

-    from fabledassistant.services.rss import _fetch_full_article
-    source = feed_title or "News"
-    content_body = (await _fetch_full_article(item.url) if item.url else None) or (item.content or "").strip()
-    seeded_text = f"**{source}**\n\n**{item.title}**"
-    if content_body:
-        seeded_text += f"\n\n{content_body}"
-    if item.url:
-        seeded_text += f"\n\nSource: {item.url}"
+    model = await get_setting(uid, "default_model", "") or Config.OLLAMA_MODEL
+    try:
+        discuss_prompt = await seed_article_discussion(conv.id, item, model)
+    except EmptyArticleError as e:
+        # Roll back the empty conversation so the user doesn't end up with a
+        # phantom entry in their chat list.
+        try:
+            await delete_conversation(uid, conv.id)
+        except Exception:
+            logger.warning("Failed to clean up empty article conversation %s", conv.id)
+        return jsonify({"error": str(e)}), 422

-    async with _async_session() as session:
-        msg = Message(
-            conversation_id=conv.id,
-            role="assistant",
-            content=seeded_text,
-            msg_metadata={"rss_item_ids": [item_id]},
-        )
-        session.add(msg)
-        await session.commit()
+    # Reload conversation so we see the two messages the helper just added.
+    conv = await get_conversation(uid, conv.id)
+    assert conv is not None

-    return jsonify({"conversation_id": conv.id}), 201
+    history: list[dict] = []
+    for msg in conv.messages:
+        if msg.role == "system":
+            continue
+        msg_dict: dict = {"role": msg.role, "content": msg.content or ""}
+        if msg.tool_calls:
+            msg_dict["tool_calls"] = [
+                {"function": {"name": tc["function"], "arguments": tc["arguments"]}}
+                for tc in msg.tool_calls
+            ]
+            history.append(msg_dict)
+            for tc in msg.tool_calls:
+                history.append({"role": "tool", "content": json.dumps(tc.get("result", {}))})
+        else:
+            history.append(msg_dict)
+
+    assistant_msg = await add_message(conv.id, "assistant", "", status="generating")
+    buf = create_buffer(conv.id, assistant_msg.id)
+    asyncio.create_task(run_generation(
+        buf, history, model, uid, conv.id, conv.title or "", discuss_prompt,
+    ))
+
+    return jsonify({
+        "conversation_id": conv.id,
+        "assistant_message_id": assistant_msg.id,
+        "status": "generating",
+    }), 202
@@ -0,0 +1,270 @@
+"""Prepare article bodies as conversation-ready context.
+
+Used by the briefing ``discuss-article`` flow and the ``/news`` discuss button.
+A raw trafilatura extraction is often too large to drop whole into a chat
+history without eating the context window, so this module runs a map-reduce
+step over oversized articles and returns a compact, structured context that
+still preserves the article's meaning across sections.
+
+Small articles pass through unchanged — map-reduce only fires when the raw
+body exceeds CHAR_BUDGET. The output is cached on ``rss_items.context_prepared``
+by the caller, so repeat discuss-clicks on the same article skip this work
+entirely.
+
+The module also owns ``seed_article_discussion``, the shared routine that
+stages a synthetic ``read_article`` tool exchange plus a conversational seed
+prompt into a conversation. Both the briefing and ``/news`` entry points call
+it so the two flows stay byte-identical — the only thing that differs between
+them is whether the conversation already existed or was freshly created.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import logging
+import re
+
+from fabledassistant.models import async_session
+from fabledassistant.models.rss_feed import RssItem
+from fabledassistant.services.chat import add_message
+from fabledassistant.services.llm import generate_completion
+
+logger = logging.getLogger(__name__)
+
+# ~12k tokens at 4 chars/token. Comfortably under OLLAMA_NUM_CTX=16384
+# with room left for system prompt, chat history, and the assistant reply.
+CHAR_BUDGET = 48_000
+
+# Chunk size for the map step on oversized articles. Overlap preserves
+# context across paragraph boundaries that happen to land mid-sentence.
+CHUNK_CHARS = 8_000
+CHUNK_OVERLAP = 400
+
+_PARA_SPLIT = re.compile(r"\n\s*\n")
+
+
+def _chunk_by_paragraph(body: str) -> list[str]:
+    """Split ``body`` into chunks of up to CHUNK_CHARS, respecting paragraphs.
+
+    Paragraphs longer than CHUNK_CHARS are split mid-paragraph as a last
+    resort. Adjacent chunks share CHUNK_OVERLAP chars of trailing text so
+    a sentence straddling the boundary stays readable on both sides.
+    """
+    paragraphs = [p.strip() for p in _PARA_SPLIT.split(body) if p.strip()]
+    chunks: list[str] = []
+    current: list[str] = []
+    current_len = 0
+    for para in paragraphs:
+        para_len = len(para)
+        if para_len > CHUNK_CHARS:
+            if current:
+                chunks.append("\n\n".join(current))
+                current, current_len = [], 0
+            for i in range(0, para_len, CHUNK_CHARS - CHUNK_OVERLAP):
+                chunks.append(para[i : i + CHUNK_CHARS])
+            continue
+        if current_len + para_len + 2 > CHUNK_CHARS and current:
+            chunks.append("\n\n".join(current))
+            tail = current[-1][-CHUNK_OVERLAP:] if current else ""
+            current = [tail, para] if tail else [para]
+            current_len = len(tail) + para_len + (2 if tail else 0)
+        else:
+            current.append(para)
+            current_len += para_len + 2
+    if current:
+        chunks.append("\n\n".join(current))
+    return chunks
+
+
+async def _summarize_chunk(title: str, chunk: str, index: int, total: int, model: str) -> str:
+    """Map-step summary of one article chunk.
+
+    Aims for ~300 words of dense, factual prose — not bullet points — so the
+    downstream chat model can quote from it naturally.
+    """
+    messages = [
+        {
+            "role": "system",
+            "content": (
+                "You are summarizing one section of a larger article so a downstream "
+                "conversation model can discuss the full article without having to read "
+                "every word.\n\n"
+                "Requirements:\n"
+                "- 250–350 words of dense factual prose\n"
+                "- Preserve specific claims, numbers, names, and quotes\n"
+                "- Do NOT editorialize or add analysis\n"
+                "- Do NOT use bullet points or headings\n"
+                "- Do NOT say 'this section' or 'this article' — write content, not meta"
+            ),
+        },
+        {
+            "role": "user",
+            "content": (
+                f"Article: {title}\n"
+                f"Section {index + 1} of {total}:\n\n{chunk}"
+            ),
+        },
+    ]
+    try:
+        # Pin num_ctx — same rationale as services/research.py:66. A large
+        # chunk plus system prompt can push well past the default window;
+        # silent truncation here would drop the tail of the chunk without
+        # any error, producing a misleading summary.
+        raw = await generate_completion(
+            messages, model, max_tokens=600, num_ctx=16384
+        )
+        return raw.strip()
+    except Exception:
+        logger.warning(
+            "Article chunk summary failed for section %d/%d of '%s'",
+            index + 1, total, title, exc_info=True,
+        )
+        # Fall back to the raw chunk truncated to ~1500 chars so the overall
+        # pipeline still delivers something rather than dropping the section.
+        return chunk[:1500]
+
+
+async def prepare_article_context(
+    title: str,
+    url: str,
+    body: str,
+    model: str,
+) -> str:
+    """Return a conversation-ready context block for ``body``.
+
+    - Small article (≤ CHAR_BUDGET): returns ``body`` unchanged.
+    - Oversized article: runs a parallel map step over paragraph-aware
+      chunks and concatenates the summaries under section headers.
+
+    The returned string is what should go into the ``read_article`` synthetic
+    tool-result in chat history. Callers are responsible for caching it to
+    ``rss_items.context_prepared``.
+    """
+    body = body or ""
+    if len(body) <= CHAR_BUDGET:
+        return body
+
+    chunks = _chunk_by_paragraph(body)
+    logger.info(
+        "Article '%s' is %d chars, map-reducing into %d chunks",
+        title, len(body), len(chunks),
+    )
+
+    summaries = await asyncio.gather(
+        *[
+            _summarize_chunk(title, chunk, i, len(chunks), model)
+            for i, chunk in enumerate(chunks)
+        ]
+    )
+
+    header = (
+        f"(This article was longer than the chat window could hold verbatim, "
+        f"so the full text was split into {len(chunks)} sections and each was "
+        "summarized below. Each section preserves specific claims, numbers, "
+        "and quotes from the original.)\n\n"
+    )
+    parts = [
+        f"## Section {i + 1}\n\n{summary}"
+        for i, summary in enumerate(summaries)
+    ]
+    return header + "\n\n".join(parts)
+
+
+# Conversational seed prompt for article discussions. Kept here so both the
+# briefing and /news entry points use the exact same wording. See
+# feedback_discuss_prompt_style memory: numbered checklists produce
+# assignment-completion responses; this conversational seed opens a dialogue.
+ARTICLE_DISCUSS_SEED = (
+    "I want to talk about this article. Start with a substantive summary "
+    "of what it's arguing and the key evidence it uses, then tell me what "
+    "stood out to you or seems worth pushing back on. I'll ask follow-ups "
+    "from there."
+)
+
+
+class EmptyArticleError(Exception):
+    """Raised when an article has no extractable body text.
+
+    Callers (the briefing and /news discuss routes) map this to a 422 so the
+    user sees a clear error instead of a hallucinated summary built from an
+    empty synthetic tool result.
+    """
+
+
+async def seed_article_discussion(
+    conv_id: int,
+    item: RssItem,
+    model: str,
+) -> str:
+    """Stage the synthetic read_article tool exchange + conversational seed.
+
+    Used by both the briefing ``discuss_article`` route and the ``/news``
+    ``from-article`` conversation creator. Handles the three-layer cache
+    (``context_prepared`` → ``content_full`` → fresh fetch) and inserts two
+    messages into ``conv_id``:
+
+    1. An assistant message with a synthetic ``read_article`` tool_call whose
+       ``result.content`` carries the prepared article context. The message
+       also carries ``msg_metadata={"rss_item_id": ...}`` so the post-generation
+       hook in ``generation_task.py`` can locate it and persist the first
+       reply as a discussion-summary Note.
+    2. A user message with the shared conversational seed prompt.
+
+    Returns the seed prompt string so callers can pass it to ``run_generation``
+    as ``user_content``.
+    """
+    # Avoid circulars: rss helper imports article_context indirectly nowhere,
+    # but keep this local for symmetry with the route-level imports it
+    # replaces.
+    from fabledassistant.services.rss import get_or_fetch_full_article
+
+    if item.context_prepared:
+        article_content = item.context_prepared
+    else:
+        raw_body = await get_or_fetch_full_article(item) or item.content or ""
+        if not raw_body.strip():
+            # Hard-fail rather than stage an empty synthetic tool result.
+            # An empty `content` field silently tells the model "the article
+            # has nothing in it" and it confabulates from RAG/history. Better
+            # to surface a clean error to the user.
+            logger.warning(
+                "Article discussion aborted: empty body for rss_item %s (%s)",
+                item.id, item.url,
+            )
+            raise EmptyArticleError(
+                "Couldn't extract any readable text from this article."
+            )
+        article_content = await prepare_article_context(
+            item.title or "", item.url, raw_body, model,
+        )
+        if not article_content.strip():
+            raise EmptyArticleError(
+                "Couldn't extract any readable text from this article."
+            )
+        async with async_session() as session:
+            fresh = await session.get(RssItem, item.id)
+            if fresh is not None:
+                fresh.context_prepared = article_content
+                await session.commit()
+
+    synthetic_tool_calls = [{
+        "function": "read_article",
+        "arguments": {"url": item.url},
+        "result": {
+            "success": True,
+            "type": "article_content",
+            "url": item.url,
+            "content": article_content,
+            "truncated": False,
+        },
+    }]
+    await add_message(
+        conv_id,
+        "assistant",
+        "",
+        status="complete",
+        tool_calls=synthetic_tool_calls,
+        msg_metadata={"rss_item_id": item.id, "article_seed": True},
+    )
+    await add_message(conv_id, "user", ARTICLE_DISCUSS_SEED)
+    return ARTICLE_DISCUSS_SEED
@@ -187,6 +187,7 @@ async def add_message(
    context_note_id: int | None = None,
    status: str | None = None,
    tool_calls: list | None = None,
+    msg_metadata: dict | None = None,
 ) -> Message:
    async with async_session() as session:
        kwargs: dict = dict(
@@ -199,6 +200,8 @@ async def add_message(
            kwargs["status"] = status
        if tool_calls is not None:
            kwargs["tool_calls"] = tool_calls
+        if msg_metadata is not None:
+            kwargs["msg_metadata"] = msg_metadata
        msg = Message(**kwargs)
        session.add(msg)
        # Touch conversation updated_at
@@ -36,71 +36,84 @@ _TOOL_CALL_MARKER = re.compile(r"^\s*\[TOOL_CALLS\]\s*", re.IGNORECASE)

 DB_FLUSH_INTERVAL = 5.0  # seconds between partial DB flushes

-# ---------------------------------------------------------------------------
-# Thinking decision
-# ---------------------------------------------------------------------------
-#
-# `_should_think` is the single source of truth for whether a qwen3-class
-# model should engage chain-of-thought for a given request. Frontend callers
-# should NOT hardcode think=True — leave it False and let the classifier
-# decide from message content. An explicit think_requested=True still acts
-# as an override for callers (e.g. a future UI toggle or MCP client) that
-# want to force extended reasoning regardless of content.
-#
-# Why gate it: on qwen3:14b, thinking adds 5–20s of latency before the first
-# visible content token, and most conversational messages do not benefit.
-# Gating by content keeps quick chats fast while preserving reasoning depth
-# for prompts that actually need it.
-#
-# Models that don't support extended reasoning (e.g. llama3, mistral) simply
-# ignore the `think` parameter in the Ollama chat request, so the decision
-# here is harmless on non-thinking models.

+async def _maybe_save_article_discussion_note(
+    user_id: int, conv_id: int, reply_content: str,
+) -> None:
+    """Persist a seeded article-discussion's first reply as a Note.

-# Keywords that strongly suggest the user wants reasoning / analysis. Matched
-# case-insensitively as whole-ish phrases.
-_THINK_KEYWORDS: tuple[str, ...] = (
-    "why", "how does", "how do i", "how would", "how should",
-    "explain", "analyze", "analyse", "compare", "contrast",
-    "design", "architect", "architecture", "plan out", "strategize",
-    "debug", "diagnose", "troubleshoot", "root cause",
-    "review", "critique", "evaluate", "trade-off", "tradeoff", "trade off",
-    "pros and cons", "step by step", "walk me through",
-    "prove", "derive", "figure out", "work through",
-    "discuss",  # covers briefing /discuss-article + /discuss-topic entry points
-)
+    Fires after ``run_generation`` completes. Looks for a synthetic
+    read_article seed message on the conversation; if found AND the linked
+    ``rss_items`` row has no ``discussion_note_id`` yet, saves ``reply_content``
+    as a Note, tags it, and writes the backlink. Subsequent discuss clicks on
+    the same article are a no-op (already linked).

-# Messages shorter than this and without any think-keyword are treated as
-# simple/conversational and skip the thinking phase.
-_SHORT_MESSAGE_CHARS = 80
-
-# Messages longer than this are treated as substantive regardless of keywords.
-_LONG_MESSAGE_CHARS = 400
-
-
-def _should_think(user_content: str, think_requested: bool) -> bool:
-    """Return whether extended thinking should be used for this request.
-
-    ``think_requested`` acts as an explicit override: if True, thinking is
-    forced on regardless of content. If False (the default), the decision is
-    made by inspecting the message: long or keyword-bearing messages get
-    thinking; short conversational messages skip it.
+    Failures are logged and swallowed — the chat UI should never break because
+    Note persistence hit a snag.
    """
-    if think_requested:
-        return True
-    text = (user_content or "").strip()
-    if not text:
-        return False
-    if len(text) >= _LONG_MESSAGE_CHARS:
-        return True
-    lowered = text.lower()
-    if any(kw in lowered for kw in _THINK_KEYWORDS):
-        return True
-    if len(text) < _SHORT_MESSAGE_CHARS:
-        return False
-    # Medium-length message with no obvious reasoning cue: default off.
-    return False
+    try:
+        if not reply_content or not reply_content.strip():
+            return
+        from sqlalchemy import select as _select
+        from fabledassistant.models.conversation import Message as _Message
+        from fabledassistant.models.rss_feed import RssItem as _RssItem
+        from fabledassistant.services.notes import create_note

+        async with async_session() as session:
+            result = await session.execute(
+                _select(_Message)
+                .where(_Message.conversation_id == conv_id)
+                .order_by(_Message.id.asc())
+            )
+            messages = result.scalars().all()
+            seed_meta = None
+            for m in messages:
+                meta = m.msg_metadata or {}
+                if meta.get("article_seed") and meta.get("rss_item_id"):
+                    seed_meta = meta
+                    break
+            if seed_meta is None:
+                return
+            item_id = int(seed_meta["rss_item_id"])
+            item = await session.get(_RssItem, item_id)
+            if item is None or item.discussion_note_id is not None:
+                return
+            article_title = (item.title or "Untitled article").strip()
+            article_url = item.url
+            article_topics = list(item.topics or [])
+
+        note_title = f"Article: {article_title}"[:200]
+        body_parts = [f"**Source:** {article_url}"] if article_url else []
+        body_parts.append(reply_content.strip())
+        note_body = "\n\n".join(body_parts)
+        tags = ["article-summary"] + [t for t in article_topics if t]
+        note = await create_note(
+            user_id=user_id,
+            title=note_title,
+            body=note_body,
+            tags=tags,
+            entity_meta={
+                "source": "article_discussion",
+                "rss_item_id": item_id,
+                "url": article_url,
+                "conversation_id": conv_id,
+            },
+        )
+
+        async with async_session() as session:
+            fresh = await session.get(_RssItem, item_id)
+            if fresh is not None and fresh.discussion_note_id is None:
+                fresh.discussion_note_id = note.id
+                await session.commit()
+        logger.info(
+            "Saved article-discussion summary as note %d for rss_item %d (conv %d)",
+            note.id, item_id, conv_id,
+        )
+    except Exception:
+        logger.warning(
+            "Failed to persist article-discussion note for conv %d",
+            conv_id, exc_info=True,
+        )

 # Human-readable labels for each tool, shown in the status indicator
 _TOOL_LABELS: dict[str, str] = {
@@ -289,11 +302,12 @@ async def run_generation(
    # Emit context event
    buf.append_event("context", {"context": context_meta})

-    # `_should_think` is authoritative — frontend callers pass think=False by
-    # default and let this classifier decide based on message content. An
-    # explicit think=True still forces on as an override.
+    # Always think on qwen3-class models: reasoning mode is the only reliable
+    # path for the tool-call template. Content-based gating was tried in 87fcaa6
+    # but exposed silent-generation failures on short tool-intent prompts, since
+    # the classifier had no way to tell that "create a task" needs a tool call.
    think_requested = think
-    think = _should_think(user_content, think_requested)
+    think = True

    t_start = time.monotonic()
    timing: dict = {
@@ -331,6 +345,14 @@ async def run_generation(
            buf.append_event("status", {"status": "Generating response..." if _round == 0 else "Composing response..."})
            t_stream = time.monotonic()

+            approx_msg_chars = sum(len(str(m.get("content", ""))) for m in messages)
+            round_content_start = len(buf.content_so_far)
+            round_output_tokens_start = timing.get("output_tokens") or 0
+            round_prompt_tokens_start = timing.get("prompt_tokens") or 0
+            logger.info(
+                "CTX_DIAG round_start conv=%d round=%d num_ctx=%d msgs=%d approx_chars=%d think=%s",
+                conv_id, _round, num_ctx, len(messages), approx_msg_chars, think,
+            )
            async for chunk in _stream_with_retry(messages, model, tools, think, num_ctx=num_ctx):
                if buf.cancel_event.is_set():
                    cancelled = True
@@ -429,6 +451,21 @@ async def run_generation(
                        all_tool_calls.append(tool_record)
                        buf.append_event("tool_call", {"tool_call": tool_record})

+            round_content_added = len(buf.content_so_far) - round_content_start
+            round_output_tokens_added = (timing.get("output_tokens") or 0) - round_output_tokens_start
+            round_prompt_tokens = (timing.get("prompt_tokens") or 0) - round_prompt_tokens_start
+            headroom = num_ctx - round_prompt_tokens if round_prompt_tokens else None
+            is_silent = (
+                not round_tool_calls
+                and round_content_added == 0
+                and round_output_tokens_added > 0
+            )
+            logger.info(
+                "CTX_DIAG round_end conv=%d round=%d think=%s prompt_tokens=%d output_tokens=%d headroom=%s content_added=%d tool_calls=%d silent=%s",
+                conv_id, _round, think, round_prompt_tokens, round_output_tokens_added,
+                headroom, round_content_added, len(round_tool_calls), is_silent,
+            )
+
            timing["generation_ms"] = int((time.monotonic() - t_stream) * 1000)

            if cancelled:
@@ -463,6 +500,29 @@ async def run_generation(
        # Strip model artifacts from final content
        buf.content_so_far = _TOOL_CALL_MARKER.sub("", buf.content_so_far)

+        # Silent-generation safety net: the model burned output tokens but
+        # nothing landed in content or tool_calls (seen with qwen3:14b when
+        # its tool-call emission doesn't parse). Show a visible fallback so
+        # the user isn't staring at an empty bubble.
+        if (
+            not cancelled
+            and not buf.content_so_far.strip()
+            and not all_tool_calls
+            and (timing.get("output_tokens") or 0) > 0
+        ):
+            logger.warning(
+                "Silent generation for conv %d: output_tokens=%s but empty content "
+                "and no tool calls (model=%s)",
+                conv_id, timing.get("output_tokens"), model,
+            )
+            fallback = (
+                "I wasn't able to produce a usable response — the model generated "
+                "tokens that couldn't be parsed as content or a tool call. "
+                "Please try rephrasing, or try again."
+            )
+            buf.content_so_far = fallback
+            buf.append_event("chunk", {"chunk": fallback})
+
        # Final save
        logger.info("Generation complete for conv %d: content_length=%d, tool_calls=%d",
                     conv_id, len(buf.content_so_far), len(all_tool_calls))
@@ -526,6 +586,18 @@ async def run_generation(
        msg_count = len(non_system)
        should_gen_title = not conv_title or (msg_count > 0 and msg_count % 10 == 0)

+        # Persist article-discussion seed conversations as a Note on their
+        # first assistant reply. This makes "Discuss" summaries part of RAG
+        # so the knowledge base stops being amnesiac about articles the user
+        # has already engaged with. The hook detects a seeded conversation by
+        # finding a synthetic read_article assistant message whose
+        # msg_metadata carries ``article_seed: True`` and whose rss_items row
+        # has no discussion_note_id yet. Fire-and-forget so the done event
+        # lands immediately.
+        asyncio.create_task(_maybe_save_article_discussion_note(
+            user_id, conv_id, buf.content_so_far,
+        ))
+
        if should_gen_title:
            # Feed the title model the *raw* conversation turns only — never
            # the post-build_context ``messages`` list. ``build_context``
@@ -265,20 +265,31 @@ async def stream_chat_with_tools(
        ) as resp:
            await _raise_ollama_error(resp, model)
            accumulated_tool_calls: list[dict] = []
+            # Silent-generation diagnostic: if Ollama reports non-zero eval_count
+            # but we never yielded any thinking/content/tool_calls, something
+            # in the frames isn't landing in a field we read. Capture the last
+            # few frames so we can see what Ollama actually sent.
+            yielded_anything = False
+            recent_frames: list[str] = []
            async for line in resp.aiter_lines():
                if not line.strip():
                    continue
+                if len(recent_frames) >= 5:
+                    recent_frames.pop(0)
+                recent_frames.append(line[:500])
                data = json.loads(line)
                msg = data.get("message", {})

                # Thinking chunks (qwen3 chain-of-thought, only when think=True)
                thinking = msg.get("thinking", "")
                if thinking:
+                    yielded_anything = True
                    yield ChatChunk(type="thinking", content=thinking)

                # Content chunks
                chunk = msg.get("content", "")
                if chunk:
+                    yielded_anything = True
                    yield ChatChunk(type="content", content=chunk)

                # Collect tool calls from any message (some models
@@ -294,13 +305,21 @@ async def stream_chat_with_tools(
                            len(accumulated_tool_calls),
                            json.dumps(accumulated_tool_calls)[:500],
                        )
+                        yielded_anything = True
                        yield ChatChunk(type="tool_calls", tool_calls=accumulated_tool_calls)
                    else:
                        logger.debug("Ollama done with no tool calls")
+                    eval_count = data.get("eval_count") or 0
+                    if not yielded_anything and eval_count > 0:
+                        logger.warning(
+                            "Ollama silent generation: model=%s eval_count=%d but no "
+                            "thinking/content/tool_calls were yielded. Last frames: %s",
+                            model, eval_count, recent_frames,
+                        )
                    yield ChatChunk(
                        type="done",
                        prompt_tokens=data.get("prompt_eval_count"),
-                        output_tokens=data.get("eval_count"),
+                        output_tokens=eval_count,
                    )
                    break

@@ -568,18 +587,26 @@ async def build_context(
        "CRITICAL: Call the tool functions directly. NEVER write out function calls as text or code. NEVER describe what you would do — just do it.",
        "GROUNDING: When the user asks about their own data — tasks, notes, events, projects, news, anything stored in this system — call the relevant tool to see what actually exists before answering. Never assert facts about the user's data from memory, prior context, or assumption. If you are unsure whether something exists, check with a tool.",
        "HONESTY WHEN EMPTY: If a tool returns empty results (no matching tasks, no events in the date range, no search hits, no notes found), tell the user plainly that nothing matched. Do not fabricate example items, do not invent plausible-sounding meetings or deadlines to fill the response, and do not hedge with generic suggestions dressed up as real data. A direct 'you don't have anything on your calendar today' is always better than an invented event.",
-        "Available actions: create_task, create_note, update_note, delete_note, delete_task, get_note, list_notes, list_tasks, search_notes.",
+    ]
+    actions = [
+        "create_note (also creates tasks — set status='todo')", "update_note", "delete_note",
+        "read_note", "list_notes", "list_tasks", "log_work", "search_notes",
+        "create_project", "list_projects", "get_project", "update_project",
+        "search_projects", "create_milestone", "update_milestone", "list_milestones",
+        "save_person", "save_place", "create_list", "add_to_list", "clear_checked_items",
+        "set_rag_scope", "get_profile", "update_profile", "get_weather", "calculate",
+        "get_rss_items", "add_rss_feed", "read_article",
    ]
    if has_caldav:
-        tool_lines[-1] = (
-            "Available actions: create_task, create_note, update_note, delete_note, delete_task, get_note, list_notes, list_tasks, search_notes, "
-            "create_event, list_events, search_events, update_event, delete_event, list_calendars."
-        )
+        actions.extend(["create_event", "list_events", "search_events", "update_event", "delete_event", "list_calendars"])
        tool_lines.append(
            "For calendar events, use ISO 8601 datetime format with the user's timezone offset (stated in context below). "
            "Always include the UTC offset in datetime strings (e.g. 2026-09-30T14:00:00+01:00)."
        )
        tool_lines.append("When the user says 'remind me' with a time before an event, use the reminder_minutes parameter.")
+    if Config.searxng_enabled():
+        actions.extend(["search_web", "research_topic", "search_images"])
+    tool_lines.append(f"Available actions: {', '.join(actions)}.")
    tool_lines.append(
        "For relative dates like 'Friday' or 'next week', resolve them to YYYY-MM-DD format. "
        "Always include the UTC offset when creating events (user's timezone is stated in context below)."
@@ -658,7 +685,7 @@ async def build_context(
                    f"\n\n--- Active Workspace ---\n"
                    f"You are in the \"{wp.title}\" project workspace.\n"
                    f"All notes and tasks you create or update MUST belong to this project.\n"
-                    f"Always pass project=\"{wp.title}\" when calling create_note or create_task.\n"
+                    f"Always pass project=\"{wp.title}\" when calling create_note.\n"
                    f"--- End Active Workspace ---"
                )
        except Exception:
@@ -724,18 +751,27 @@ async def build_context(
    orphan_only = rag_project_id is None
    effective_project_id = rag_project_id if (rag_project_id is not None and rag_project_id != -1) else None

-    try:
-        from fabledassistant.services.embeddings import semantic_search_notes
-        for score, note in await semantic_search_notes(
-            user_id, user_message, exclude_ids=search_exclude or None, limit=8,
-            project_id=effective_project_id,
-            orphan_only=orphan_only,
-        ):
-            found_scored.append((score, note))
-    except Exception:
-        logger.warning("Semantic note search failed, falling back to keyword search", exc_info=True)
+    # Skip RAG auto-injection on the first turn of a seeded article discussion.
+    # The article body is already the sole context the user wants — pulling in
+    # unrelated orphan notes tricks the model into summarizing those instead.
+    # Follow-up turns keep RAG on because by then the user's own messages drive
+    # the query rather than the generic seed prompt.
+    from fabledassistant.services.article_context import ARTICLE_DISCUSS_SEED
+    _skip_rag_for_article_seed = user_message.strip() == ARTICLE_DISCUSS_SEED

-    if not found_scored:
+    if not _skip_rag_for_article_seed:
+        try:
+            from fabledassistant.services.embeddings import semantic_search_notes
+            for score, note in await semantic_search_notes(
+                user_id, user_message, exclude_ids=search_exclude or None, limit=8,
+                project_id=effective_project_id,
+                orphan_only=orphan_only,
+            ):
+                found_scored.append((score, note))
+        except Exception:
+            logger.warning("Semantic note search failed, falling back to keyword search", exc_info=True)
+
+    if not found_scored and not _skip_rag_for_article_seed:
        keywords = _extract_keywords(user_message)
        if keywords:
            try:
@@ -34,6 +34,34 @@ def _html_to_text(html: str) -> str:
        return html


+async def get_or_fetch_full_article(item: RssItem) -> str | None:
+    """Return the full article body, fetching+caching on miss.
+
+    Checks ``item.content_full`` first — populated either by the enrichment
+    pass at feed-ingest time or by a previous discuss-click. On miss, fetches
+    via ``_fetch_full_article`` and writes through. Returns ``None`` only if
+    the fetch itself fails; ``item.content_full == ""`` is still a cache hit.
+
+    Callers must pass an RssItem attached to an open session if they want
+    the write-through to persist — otherwise the fetched text is returned
+    but the cache stays empty and the next click will re-fetch.
+    """
+    if item.content_full is not None:
+        return item.content_full
+    if not item.url:
+        return None
+    text = await _fetch_full_article(item.url)
+    if text is None:
+        return None
+    async with async_session() as session:
+        fresh = await session.get(RssItem, item.id)
+        if fresh is not None:
+            fresh.content_full = text
+            fresh.content_fetched_at = datetime.now(timezone.utc)
+            await session.commit()
+    return text
+
+
 async def _fetch_full_article(url: str) -> str | None:
    """Fetch a URL and extract its main article text via trafilatura.

@@ -209,6 +237,11 @@ async def fetch_and_cache_feed(feed_id: int, url: str) -> int:
                        item = await session.get(RssItem, item_id)
                        if item:
                            item.content = full_text
+                            # Populate the discuss-click cache too so the
+                            # first click skips straight to the map-reduce
+                            # step without re-fetching.
+                            item.content_full = full_text
+                            item.content_fetched_at = datetime.now(timezone.utc)
                            await session.commit()
                            await upsert_rss_item_embedding(
                                item_id, feed_user_id, item.title or "", item.content
@@ -134,3 +134,83 @@ def test_history_builder_no_tool_calls_unchanged():
    assert len(history) == 2
    assert history[0] == {"role": "user", "content": "Hello"}
    assert history[1] == {"role": "assistant", "content": "Hi there!"}
+
+
+# ---------------------------------------------------------------------------
+# prepare_article_context tests
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.asyncio
+async def test_prepare_article_context_small_passthrough():
+    """Articles under CHAR_BUDGET pass through unchanged with zero LLM calls."""
+    from fabledassistant.services import article_context
+
+    body = "A short article.\n\nWith two paragraphs."
+    with patch(
+        "fabledassistant.services.article_context.generate_completion",
+        new_callable=AsyncMock,
+    ) as mock_gen:
+        out = await article_context.prepare_article_context(
+            "Title", "https://example.com", body, "test-model",
+        )
+
+    assert out == body
+    mock_gen.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_prepare_article_context_large_runs_map_reduce():
+    """Articles over CHAR_BUDGET are chunked and map-reduced via the background model."""
+    from fabledassistant.services import article_context
+
+    # CHAR_BUDGET is 48_000 — build a body well over that with paragraph breaks
+    # so the chunker has natural splits to work with.
+    paragraph = ("Lorem ipsum dolor sit amet, consectetur adipiscing elit. " * 40).strip()
+    body = "\n\n".join([paragraph] * 30)  # ~70k+ chars across 30 paragraphs
+    assert len(body) > article_context.CHAR_BUDGET
+
+    with patch(
+        "fabledassistant.services.article_context.generate_completion",
+        new_callable=AsyncMock,
+        return_value="Summary of this section with specific claims preserved.",
+    ) as mock_gen:
+        out = await article_context.prepare_article_context(
+            "Long Article", "https://example.com/long", body, "test-model",
+        )
+
+    # At least one LLM call fired (the map step ran)
+    assert mock_gen.await_count >= 1
+    # Output carries the oversized-article header and section markers
+    assert "longer than the chat window" in out
+    assert "## Section 1" in out
+    # Map output is much smaller than the raw body
+    assert len(out) < len(body)
+
+
+def test_chunk_by_paragraph_respects_boundaries():
+    """Chunker splits on paragraph breaks, not mid-sentence."""
+    from fabledassistant.services.article_context import _chunk_by_paragraph, CHUNK_CHARS
+
+    paragraphs = [f"Paragraph {i}. " + ("x" * 1000) for i in range(20)]
+    body = "\n\n".join(paragraphs)
+
+    chunks = _chunk_by_paragraph(body)
+
+    # Each chunk stays under the budget
+    for chunk in chunks:
+        assert len(chunk) <= CHUNK_CHARS
+    # Total content is preserved (modulo overlap duplication, so ≥ original)
+    assert sum(len(c) for c in chunks) >= len(body) * 0.9
+
+
+def test_chunk_by_paragraph_handles_oversized_paragraph():
+    """A single paragraph larger than CHUNK_CHARS gets split mid-paragraph."""
+    from fabledassistant.services.article_context import _chunk_by_paragraph, CHUNK_CHARS
+
+    body = "x" * (CHUNK_CHARS * 3)  # one huge paragraph, no breaks
+    chunks = _chunk_by_paragraph(body)
+
+    assert len(chunks) >= 3
+    for chunk in chunks:
+        assert len(chunk) <= CHUNK_CHARS
@@ -1,127 +0,0 @@
-"""Tests for the `_should_think` classifier.
-
-`_should_think` decides whether qwen3-class models should engage chain-of-
-thought for a given chat turn. It is the single source of truth: frontend
-callers pass `think_requested=False` by default and defer to this function,
-while explicit `think_requested=True` acts as an override for curated
-analytical entry points.
-
-These tests lock in the content-based behavior so future tweaks don't
-silently regress the short / long / keyword boundaries.
-"""
-
-import pytest
-
-from fabledassistant.services.generation_task import (
-    _LONG_MESSAGE_CHARS,
-    _SHORT_MESSAGE_CHARS,
-    _should_think,
-)
-
-
-class TestExplicitOverride:
-    def test_override_forces_on_for_empty(self):
-        assert _should_think("", think_requested=True) is True
-
-    def test_override_forces_on_for_short_greeting(self):
-        assert _should_think("hi", think_requested=True) is True
-
-    def test_override_forces_on_for_medium_no_keyword(self):
-        text = "just checking in on the status of things for the week"
-        assert _should_think(text, think_requested=True) is True
-
-
-class TestEmptyAndWhitespace:
-    def test_empty_string_off(self):
-        assert _should_think("", think_requested=False) is False
-
-    def test_none_content_off(self):
-        # _should_think defensively handles None content from upstream callers
-        assert _should_think(None, think_requested=False) is False  # type: ignore[arg-type]
-
-    def test_whitespace_only_off(self):
-        assert _should_think("   \n\t  ", think_requested=False) is False
-
-
-class TestShortMessages:
-    def test_short_greeting_off(self):
-        assert _should_think("hi", think_requested=False) is False
-
-    def test_short_thanks_off(self):
-        assert _should_think("thanks!", think_requested=False) is False
-
-    def test_short_acknowledgement_off(self):
-        assert _should_think("ok sounds good", think_requested=False) is False
-
-    def test_just_below_short_threshold_off(self):
-        text = "a" * (_SHORT_MESSAGE_CHARS - 1)
-        assert _should_think(text, think_requested=False) is False
-
-
-class TestLongMessages:
-    def test_at_long_threshold_on(self):
-        text = "a" * _LONG_MESSAGE_CHARS
-        assert _should_think(text, think_requested=False) is True
-
-    def test_well_above_long_threshold_on(self):
-        text = "x" * (_LONG_MESSAGE_CHARS * 3)
-        assert _should_think(text, think_requested=False) is True
-
-
-class TestMediumMessages:
-    def test_medium_no_keyword_off(self):
-        # Between the short and long thresholds with no reasoning cue.
-        text = "a" * ((_SHORT_MESSAGE_CHARS + _LONG_MESSAGE_CHARS) // 2)
-        assert _should_think(text, think_requested=False) is False
-
-
-class TestKeywordTriggers:
-    @pytest.mark.parametrize(
-        "text",
-        [
-            "why is this failing",
-            "how does caching work here",
-            "how do i configure this",
-            "explain the retry logic",
-            "analyze the latency breakdown",
-            "compare gemma3 vs qwen3 for tool use",
-            "please design the schema for X",
-            "debug this error",
-            "troubleshoot the connection issue",
-            "root cause the outage",
-            "review this PR",
-            "critique my approach",
-            "walk me through the flow",
-            "step by step instructions please",
-            "pros and cons of each option",
-            "help me figure out what's wrong",
-            "discuss this article",  # covers briefing /discuss entry points
-        ],
-    )
-    def test_keyword_forces_on(self, text):
-        assert _should_think(text, think_requested=False) is True
-
-    def test_keyword_case_insensitive(self):
-        assert _should_think("WHY does this break?", think_requested=False) is True
-
-    def test_keyword_in_longer_sentence(self):
-        text = "hey quick one — can you explain what caching does for qwen3"
-        assert _should_think(text, think_requested=False) is True
-
-
-class TestNonTriggers:
-    """Messages that look chatty and should NOT trigger thinking."""
-
-    @pytest.mark.parametrize(
-        "text",
-        [
-            "hey",
-            "yep",
-            "no worries",
-            "got it, thanks",
-            "good morning",
-            "remind me later",  # no reasoning keyword, short
-        ],
-    )
-    def test_chatty_messages_off(self, text):
-        assert _should_think(text, think_requested=False) is False
Author	SHA1	Message	Date
bvandeusen	72708475a3	Release v26.04.16.1 — Fix tool actions mismatch in system prompt	2026-04-16 12:39:02 +00:00
bvandeusen	e07d8436b7	fix(llm): sync available actions list with actual registered tools The system prompt listed phantom tools (create_task, delete_task, get_note) that don't exist, causing the model to spiral when users asked to create tasks under a project. Replaced the stale hardcoded string with a dynamically-built actions list matching all registered tools, and added conditional searxng/caldav extensions. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-04-16 08:16:36 -04:00
bvandeusen	1711ee6a1f	Release v26.04.15.1 — Chat UI overhaul + silent gen fix	2026-04-16 05:52:47 +00:00
bvandeusen	369cf0a7c9	fix(chat): default context sidebar to hidden Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-04-15 23:29:14 -04:00
bvandeusen	3f2813d16f	feat(chat): Phase D empty-state + overlay context sidebar Empty-state rethink: replace "Start a conversation." with a richer landing showing recent conversations (top 5 as resume links) and a voice-mode entry button. Gated to full variant only; widget/briefing keep the simple message. Context sidebar: switch from grid column (4-col layout) to absolute overlay on the right gutter (3-col layout). The reading column is now always symmetrically centered regardless of whether the sidebar is visible — eliminates the leftward shift that occurred when context notes appeared. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-04-15 22:35:39 -04:00
bvandeusen	fddac2aa2f	fix(chat): always think on qwen3, drop content-based classifier Content-based gating (_should_think) was introduced in `87fcaa6` to cut TTFT on simple prompts, but it has no way to tell that short prompts like "create a task titled X" are going to trigger a tool call — and qwen3:14b's tool-call template is unreliable at think=False, producing intermittent silent generations where output tokens burn but nothing parses into content or tool_calls. Reverting to always-on thinking restores the pre-87fcaa6 reliability of tool emission at the cost of TTFT latency on short conversational prompts. This also lets us delete the silent-round retry loop (which can no longer fire) along with its bookkeeping. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-04-15 21:09:16 -04:00
bvandeusen	1261e93ede	chore(generation): track CTX_DIAG per attempt not per round Retry attempts were previously conflated with the initial call, making prompt_tokens and headroom look cumulative and useless for diagnosing the silent-round behavior. Move start-of-attempt captures inside the retry loop and emit attempt_start / attempt_end lines so each attempt's numbers stand alone. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-04-15 20:39:49 -04:00
bvandeusen	b6165e56e5	chore(generation): add CTX_DIAG logs for silent-round investigation Log num_ctx, message count, prompt/output tokens, headroom, and a silent flag per round so we can correlate silent generations against context pressure on the dev instance. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-04-15 20:09:08 -04:00
bvandeusen	5e281c534a	fix(chat): add bottom padding to messages container to clear input bar Streaming bubble (Generating response…) was flush against the input wrapper top border. Give the scroll column 0.75rem of breathing room so the assistant bubble sits visibly separate from the input. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-04-15 19:29:14 -04:00
bvandeusen	9082281225	feat(chat): Phase C — collapsible context sidebar with header toggle Context sidebar sections (Auto-included / Suggested / In Context) are now collapsible with chevrons, and collapsed state persists per conversation via localStorage. A new Context (N) button in the chat header toggles the whole sidebar — open by default on wide screens (>=1200px), closed by default below. Visual differentiation: auto- included notes get a muted background with an AUTO pill, in-context notes get an active-chip style with a primary-colored left border, and each section header shows its item count. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-04-15 19:12:11 -04:00
bvandeusen	058d6089b1	fix(chat): retry silent rounds with think=True before falling back Qwen3's tool-call tokens sometimes fail to parse into either content or tool_calls, burning output tokens and producing empty bubbles. Detect the signature within a round (empty content, no tool calls, eval_count > 0) and re-run the same round once with reasoning mode enabled, which emits more reliable output. The post-loop fallback remains as the final catch. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-04-15 18:22:28 -04:00
bvandeusen	ba0cb07c91	fix(chat): surface silent generations instead of empty bubbles Qwen3:14b sometimes burns output tokens on tool-calling attempts whose emission doesn't parse into any field we read — eval_count > 0 but no thinking/content/tool_calls ever stream to the caller. Generation completes "successfully," the user sees an empty assistant bubble, and no error is logged. Seen in conv 220 today. Two safety rails: - stream_chat_with_tools now tracks whether it yielded anything; when Ollama's done frame reports eval_count > 0 with zero yields, log a warning including the last ~5 raw frames so the next occurrence leaves breadcrumbs for diagnosis. - run_generation checks the same post-condition after the tool loop exits and, if content is empty with no tool calls but output_tokens > 0, substitutes a visible fallback message and streams it as a chunk so the user gets something readable instead of a blank bubble. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-04-15 17:35:30 -04:00
bvandeusen	4228e9a384	feat(chat): move scope chip to header, consolidate voice controls into input bar Phase B of chat view refinement. Bumps chat reading column to 1200px, lifts the RAG scope chip out of the footer and into the ChatView header (next to the title), and collapses the listen toggle / volume slider / stop-playback buttons into a single speaker popover on the input bar. Listen mode + TTS are now available in briefing and workspace chats too, since the controls live on the shared ChatInputBar. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-04-15 09:44:16 -04:00
bvandeusen	035ec0c0dc	feat(chat): Phase A chrome cleanup — kebabs, search, full-width new chat Header: - Remove Research button + modal (skill fires reliably from normal chat) - Move Summarize-as-Note into a header kebab (disabled until the conversation has messages) Left sidebar: - Full-width "+ New Chat" primary button - Search input filters conversations by title (case-insensitive substring) with a "no matches" empty state - Move Select into a sidebar kebab next to search Shared .btn-kebab + .kebab-menu styles give future conv-level actions a home. Document mousedown + Escape dismiss both kebabs. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-04-15 09:28:48 -04:00
bvandeusen	f0c93ffa3b	fix(chat): drop display:flex from .chat-panel-fill so grid layout wins The inherited .chat-panel-fill class lands on ChatPanel's new .chat-full root via attribute inheritance, and its display:flex was overriding the grid layout — stacking messages, sidebar, and input bar vertically inside the reading column instead of placing the sidebar in grid col 4. Layout is ChatPanel's responsibility now; .chat-panel-fill only sizes. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-04-15 07:40:44 -04:00
bvandeusen	ad4fe3d783	feat(chat): center chat reading column and full-height context sidebar Replaces the full-variant flex layout with a single CSS grid owning the messages column, input bar, and context sidebar. Messages and input bar share a centered ~820px reading track (--chat-reading-width token) so the mic button can't get pushed off-screen on wide monitors, and the context sidebar spans grid-row 1/-1 so it runs the full height from the chat header down to the bottom edge instead of stopping above the input bar. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-04-15 07:11:08 -04:00
bvandeusen	dcbe018297	Merge pull request 'Release v26.04.15.1 — Dynamic voice silence threshold + filled mic halo' (#35 ) from dev into main	2026-04-15 04:38:27 +00:00
bvandeusen	36fb71699b	feat(voice): dynamic silence threshold + filled red mic halo The web silence detector previously ran RMS over getByteFrequencyData bytes (which are already dB-scaled) and re-log'd the result, producing numbers that didn't line up with real dBFS. Combined with a static -40 dB threshold that sat right on top of room ambient, silence detection rarely fired and the user had to click stop manually. - Rewrite useSilenceDetector to use getFloatTimeDomainData for honest linear RMS → dBFS. - Silence threshold is now dynamic: track session peak dBFS and treat "silent" as 15 dB below peak. Auto-calibrates per mic/room. - Grace period (1500 ms) at start so the user can begin speaking before checks arm; static -45 dB fallback until peak clears -25 dB so dead-silent sessions don't spin forever. - Silence duration bumped 1500 → 2000 ms for breathing room. Visual: detached red radial-gradient disc sits behind the mic in a wrapper; the button no longer scales, so the white mic icon stays legible on top while the halo grows from 1x to ~2.6x with live amplitude. Much more obvious "you are live" signal than the prior subtle box-shadow pulse. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-04-15 00:26:52 -04:00
bvandeusen	027fbec606	Merge pull request 'Release v26.04.14.3 — Live amplitude mic pulse' (#34 ) from dev into main	2026-04-15 02:54:42 +00:00
bvandeusen	730dbfaf7b	feat(voice): pulse mic button with live amplitude The mic button now scales and glows proportional to the silence detector's RMS amplitude instead of sitting static. Gives obvious feedback that audio is actually being picked up — silence still breathes via a 0.1 floor, loud input caps at ~1.18x scale so it doesn't punch through the input bar. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-04-14 22:46:37 -04:00
bvandeusen	267a975455	Merge pull request 'Release v26.04.14.2 — Discuss failure mode fixes' (#33 ) from dev into main	2026-04-15 01:45:36 +00:00
bvandeusen	7bd1548f71	fix(discuss): hard-fail empty articles and skip RAG on seed turn Discuss flow was hallucinating unrelated content when article extraction returned empty or RAG pulled in orphan notes that looked more relevant than the generic seed prompt. - seed_article_discussion raises EmptyArticleError on empty body; briefing and /news routes return 422 instead of staging an empty synthetic tool result. - build_context skips RAG auto-injection when user_message matches ARTICLE_DISCUSS_SEED so the article IS the context on turn one; follow-up turns keep RAG on. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-04-14 18:13:17 -04:00
bvandeusen	9f3b3450fa	Merge pull request 'Release v26.04.14.1' (#32 ) from dev into main	2026-04-14 12:02:13 +00:00
bvandeusen	ba90ad8132	feat(article-discuss): unify /news + briefing entry points, persist summaries to RAG Both the /news discuss button and the briefing discuss button now call a shared seed_article_discussion() helper that stages the synthetic read_article tool exchange and the conversational seed prompt — behavior stays byte-identical across entry points. /news also auto-starts generation so the chat screen lands on an in-flight stream. First assistant reply in a seeded article conversation is persisted as a Note (tags: article-summary + article topics) and backlinked via rss_items.discussion_note_id, so the knowledge base stops being amnesiac about articles the user has engaged with. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-04-14 07:54:24 -04:00
bvandeusen	9157740069	ci: gate typecheck/lint/test on ref so main merge commits skip work Forgejo Actions doesn't consistently honor `on.push.branches` as a filter — the merge commit landing on main after a release PR was still triggering the full CI workflow on a SHA that had already passed on dev, burning ~1 minute of runner time for no reason. The `build` job already had a ref guard, so no duplicate images were being pushed, but typecheck/lint/test ran unconditionally. Add the same guard (`refs/heads/dev` or `refs/tags/v*`) to those three so main pushes trigger the workflow but every job skips immediately with zero runner time. Also tightened the build job's tag filter from `refs/tags/` to `refs/tags/v` for consistency with the new guards and to avoid ever building from a non-version tag. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-04-13 23:40:41 -04:00
bvandeusen	fd885c1bc8	Merge pull request 'Release v26.04.13.3' (#31 ) from dev into main	2026-04-14 03:36:09 +00:00
bvandeusen	8205590f8d	feat(briefing): cache + map-reduce article context for rich discuss chats The Discuss button on news cards was producing one-shot replies because the model got the whole trafilatura blob dropped into history with a canned "summarize and discuss this article" prompt — no length guard, no prep, no invitation to converse. Large articles got silently truncated by Ollama; small articles got a tepid reply. This reworks discuss_article around a three-layer cache: context_prepared → content_full → fresh trafilatura fetch First click on a small article fetches once, writes through to both caches, and passes the body straight into the synthetic read_article tool-result. First click on a large article additionally runs a parallel map step (services/article_context.py) that chunks the body on paragraph boundaries, summarizes each ~8k chunk to ~300 words of dense factual prose via the background model, and concatenates the summaries under section headers — all pinned to num_ctx=16384 so the map step doesn't itself fall victim to silent truncation. Repeat clicks on either path skip straight to the chat turn. The canned summary prompt is replaced with a conversational seed that invites the user into an actual discussion rather than a one-shot synopsis, matching the goal of "have a conversation about an article, not just read it." discuss_topic is intentionally left untouched — it's the multi-article aggregation path and needs a separate rework. Follow-up task will decide whether to retire it or rework it on the cached-context approach. Closes task #106. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-04-13 20:52:00 -04:00