feat: stream Reginald's replies token-by-token

The 26b is slow (~30s/call); the chat now shows the answer forming instead of
freezing until it's done. The final prose streams over SSE; tool-calling turns
stay structured (no partial tokens), so streaming kicks in for the narration.

core (@commitea/core):
- chat-client.complete gains an optional onToken — when set, it requests
  stream:true and parses the OpenAI SSE stream, emitting content deltas and
  assembling streamed tool-call argument fragments into the final result.
- GiteaHttpResponse exposes the optional `body` stream (real fetch has it; stubs
  don't). agent-loop threads onToken to each completion.

app:
- model:chat forwards each delta to the renderer (event.sender.send); preload
  exposes model.onToken(cb) → unsubscribe. useChat accumulates the live stream
  into a growing bubble (with a cursor), replaced by the authoritative final
  content when the turn resolves. Unconfigured → scripted reply, unchanged.

Verified: 118 core tests green (2 streaming: SSE content deltas + tool-call
fragment assembly), desktop typecheck clean, 14 fixture e2e green. Live: a real
turn against gemma-4-26b assembles the correct answer via the streaming path
(live-reginald green) — the reply now renders token-by-token.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

apps/desktop/src/main/model.ts

@@ -86,11 +86,15 @@ export function registerModelIpc(): void {
     return { configured: true, model }
   })
 
-  ipcMain.handle('model:chat', async (_event, messages: ChatMessage[]) => {
+  ipcMain.handle('model:chat', async (event, messages: ChatMessage[]) => {
     if (!router) return { ok: false as const, reason: 'unconfigured' as const }
     const client = getGiteaClient()
     const model = await resolveLoadedModel(router.small.baseUrl, router.small.model)
     const chat = createChatClient({ ...router.small, model }, fetch)
+    // stream the model's prose to the renderer token-by-token
+    const onToken = (delta: string) => {
+      if (!event.sender.isDestroyed()) event.sender.send('model:chat:token', delta)
+    }
 
     // Proposals the model formulates this turn; the renderer approves them (the
     // write happens through gitea:applyChange, never inside the loop).
@@ -129,10 +133,11 @@ export function registerModelIpc(): void {
 
     try {
       const turn = await runAgentTurn({
-        complete: (m, t) => chat.complete(m, t),
+        complete: (m, t, ot) => chat.complete(m, t, ot),
         messages: [{ role: 'system', content: REGINALD_SYSTEM }, ...messages],
         tools: REGINALD_TOOLS,
         execute,
+        onToken,
       })
       return { ok: true as const, content: turn.content, steps: turn.steps, proposals }
     } catch (e) {

apps/desktop/src/preload/index.ts

@@ -25,6 +25,12 @@ const api = {
     status: () => ipcRenderer.invoke('model:status'),
     /** One agent turn: messages in, Reginald's prose + the tools it consulted out. */
     chat: (messages: unknown) => ipcRenderer.invoke('model:chat', messages),
+    /** Subscribe to streamed prose tokens for the in-flight turn; returns an unsubscribe. */
+    onToken: (cb: (delta: string) => void) => {
+      const listener = (_e: unknown, delta: string) => cb(delta)
+      ipcRenderer.on('model:chat:token', listener)
+      return () => ipcRenderer.removeListener('model:chat:token', listener)
+    },
     /** Decompose a braindump into a proposed issue set (capture_work). */
     capture: (braindump: string) => ipcRenderer.invoke('model:capture', braindump),
   },

apps/desktop/src/renderer/src/components/shell/chat-panel.tsx

@@ -19,7 +19,7 @@ export interface ChatPanelProps {
 }
 
 export function ChatPanel({ onOpenDirectives, offline, onApplyChange }: ChatPanelProps) {
-  const { msgs, thinking, live, model, steps, proposals, send: sendChat, approve, dismiss } = useChat(onApplyChange)
+  const { msgs, thinking, live, model, steps, proposals, streaming, send: sendChat, approve, dismiss } = useChat(onApplyChange)
   // shorten "google/gemma-4-26b-a4b-qat" → "gemma-4-26b" for the header chip
   const modelLabel = model ? (model.split('/').pop() ?? model).replace(/-(qat|instruct|it|gguf)$/i, '') : 'gemma-4'
   const [text, setText] = useState('')
@@ -28,7 +28,7 @@ export function ChatPanel({ onOpenDirectives, offline, onApplyChange }: ChatPane
   useEffect(() => {
     const el = scrollRef.current
     if (el) el.scrollTop = el.scrollHeight
-  }, [msgs, thinking])
+  }, [msgs, thinking, streaming])
 
   const send = () => {
     const t = text.trim()
@@ -98,7 +98,14 @@ export function ChatPanel({ onOpenDirectives, offline, onApplyChange }: ChatPane
             The model is away from its desk. Reads still work; writes will wait their turn.
           </div>
         ) : null}
-        {thinking ? <div style={{ font: 'var(--text-agent)', color: 'var(--ink-3)' }}>considering…</div> : null}
+        {streaming ? (
+          <div style={{ font: 'var(--text-agent)', color: 'var(--ink-1)', lineHeight: 1.55 }}>
+            {streaming}
+            <span style={{ opacity: 0.5 }}>▊</span>
+          </div>
+        ) : thinking ? (
+          <div style={{ font: 'var(--text-agent)', color: 'var(--ink-3)' }}>considering…</div>
+        ) : null}
         {!thinking && steps.length ? (
           <div style={{ font: 'var(--text-caption)', color: 'var(--ink-3)', display: 'flex', alignItems: 'center', gap: 5 }}>
             <Icon name="eye" size={11} /> consulted {Array.from(new Set(steps.map((s) => s.replace('query_project', 'the project').replace('propose_change', 'the labels').replace('record_directive', 'the directive ledger')))).join(', ')}

apps/desktop/src/renderer/src/global.d.ts

@@ -68,6 +68,8 @@ export interface ModelBridge {
   status(): Promise<{ configured: boolean; model: string | null }>
   chat(messages: ChatMessage[]): Promise<ChatResult>
   capture(braindump: string): Promise<CaptureResult>
+  /** Subscribe to streamed prose tokens; returns an unsubscribe fn. */
+  onToken(cb: (delta: string) => void): () => void
 }
 
 /** The result of reading the directive ledger. */

apps/desktop/src/renderer/src/lib/use-chat.ts

@@ -20,6 +20,8 @@ export interface ChatState {
   steps: string[]
   /** Changes Reginald has proposed and is awaiting approval on. */
   proposals: ChangeProposal[]
+  /** The in-flight streamed prose (grows token-by-token) before the turn finalizes. */
+  streaming: string
   send: (text: string) => void
   approve: (p: ChangeProposal) => void
   dismiss: (p: ChangeProposal) => void
@@ -41,6 +43,7 @@ export function useChat(onApplyChange?: (change: IssueChange) => Promise<{ ok: b
   const [model, setModel] = useState<string | null>(null)
   const [steps, setSteps] = useState<string[]>([])
   const [proposals, setProposals] = useState<ChangeProposal[]>([])
+  const [streaming, setStreaming] = useState('')
   const convoRef = useRef(convo)
   convoRef.current = convo
 
@@ -86,10 +89,17 @@ export function useChat(onApplyChange?: (change: IssueChange) => Promise<{ ok: b
         role: m.from === 'user' ? 'user' : 'assistant',
         content: m.text,
       }))
+      setStreaming('')
+      const unsubscribe = window.commitea.model.onToken((delta) => setStreaming((s) => s + delta))
+      const finish = () => {
+        unsubscribe()
+        setThinking(false)
+        setStreaming('')
+      }
       window.commitea.model
         .chat(wire)
         .then((res) => {
-          setThinking(false)
+          finish()
           if (res.ok) {
             setSteps(res.steps.map((s) => s.tool))
             setProposals(res.proposals)
@@ -105,7 +115,7 @@ export function useChat(onApplyChange?: (change: IssueChange) => Promise<{ ok: b
           }
         })
         .catch(() => {
-          setThinking(false)
+          finish()
           setConvo((c) => [...c, { from: 'agent', text: 'I could not reach the model.' }])
         })
     },
@@ -137,5 +147,5 @@ export function useChat(onApplyChange?: (change: IssueChange) => Promise<{ ok: b
     setConvo((c) => [...c, { from: 'agent', text: `Left #${p.change.issue} as it was.` }])
   }, [])
 
-  return { msgs: [...seed, ...convo], thinking, live, model, steps, proposals, send, approve, dismiss }
+  return { msgs: [...seed, ...convo], thinking, live, model, steps, proposals, streaming, send, approve, dismiss }
 }