Assistant streams its answer token by token

ai/client.py streams the completion and puts the chunks back together
(Assembler), covering OpenAI's piecewise tool-call arguments, Ollama's whole
calls, OpenRouter's comment lines, unnumbered calls and in-stream errors.
The turn emits a delta event per piece and a message per round as before,
and the chat grows the last answer as deltas arrive.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WgBfkm4rCmtd1T4vHWjEPx

frontend/src/client/sdk.gen.ts

@@ -115,8 +115,9 @@ export class AiService {
      * Chat
      * One turn, streamed as server-sent events.
      *
-     * ``message``, ``tool_call`` and ``tool_result`` events as the model works,
-     * ``error`` when something stops it, and ``done`` last either way.
+     * ``delta`` as the model writes, ``message`` with a round's whole text,
+     * ``tool_call`` and ``tool_result`` as it works, ``error`` when something
+     * stops it, and ``done`` last either way.
      * @param data The data for the request.
      * @param data.requestBody
      * @returns unknown Successful Response

frontend/src/components/Chat/ChatPanel.tsx

@@ -175,10 +175,12 @@ export function ChatPanel() {
       .finally(() => setLoaded(true))
   }, [open, loaded, available])
 
-  // biome-ignore lint/correctness/useExhaustiveDependencies: scroll when the list grows, not on every render.
+  const tail = shown[shown.length - 1]
+  const tailLength = tail && "text" in tail ? tail.text.length : 0
+  // biome-ignore lint/correctness/useExhaustiveDependencies: scroll when the list or a streamed answer grows, not on every render.
   useEffect(() => {
     bottom.current?.scrollIntoView({ block: "end" })
-  }, [shown.length, waiting])
+  }, [shown.length, tailLength, waiting])
 
   const send = async () => {
     const text = draft.trim()
@@ -190,11 +192,18 @@ export function ChatPanel() {
     controller.current = abort
     const onEvent = (event: ChatEvent) => {
       switch (event.type) {
+        // A round's text grows in the last item and is settled by `message`;
+        // a tool call or the next turn always comes between two rounds.
+        case "delta":
         case "message":
-          setItems((prev) => [
-            ...prev,
-            { kind: "assistant", text: event.content },
-          ])
+          setItems((prev) => {
+            const last = prev[prev.length - 1]
+            if (last?.kind !== "assistant")
+              return [...prev, { kind: "assistant", text: event.content }]
+            const text =
+              event.type === "delta" ? last.text + event.content : event.content
+            return [...prev.slice(0, -1), { kind: "assistant", text }]
+          })
           break
         case "tool_call":
           setItems((prev) => [

frontend/src/components/Chat/stream.ts

@@ -3,6 +3,7 @@ import { apiToken } from "@/lib/portal"
 
 export type ChatEvent =
   | { type: "start" }
+  | { type: "delta"; content: string }
   | { type: "message"; content: string }
   | { type: "tool_call"; name: string; arguments: Record<string, unknown> }
   | { type: "tool_result"; name: string; ok: boolean; error?: string }