Summarize streams with real progress: adapters reassemble SSE/NDJSON deltas and report 0-99% as tokens arrive

backend/shonar/services/ai/ollama.py

@@ -8,7 +8,12 @@ from contextlib import asynccontextmanager
 import httpx
 
 from shonar.services.ai import ProviderConfigError, ProviderTransientError, SummaryResult
-from shonar.services.ai._llm import SYSTEM_PROMPT, build_user_message, parse_summary
+from shonar.services.ai._llm import (
+    SYSTEM_PROMPT,
+    build_user_message,
+    make_progress_ticker,
+    parse_summary,
+)
 
 
 class OllamaProvider:
@@ -39,10 +44,13 @@ class OllamaProvider:
                 yield client
 
     async def summarize(self, transcript: str, *, title: str | None = None,
-                        tone: str | None = None) -> SummaryResult:
+                        tone: str | None = None, on_progress=None) -> SummaryResult:
+        ticker = make_progress_ticker(on_progress)
         payload = {
             "model": self.model,
-            "stream": False,
+            # Stream (NDJSON lines) so the UI shows real progress while the
+            # model generates; the reply is reassembled from the chunks.
+            "stream": True,
             "format": "json",
             # qwen3-family models "think" by default: a long reasoning
             # chain before the JSON answer, brutally slow on CPU and it
@@ -57,20 +65,20 @@ class OllamaProvider:
             ],
         }
         try:
-            async with self._client() as client:
-                resp = await client.post(f"{self.base_url}/api/chat", json=payload)
+            async with self._client() as client, client.stream(
+                "POST", f"{self.base_url}/api/chat", json=payload
+            ) as resp:
+                if resp.status_code == 404:
+                    # Missing model and missing route both 404 here; both
+                    # are configuration, not weather.
+                    raise ProviderConfigError(
+                        "Ollama has no such model or route (HTTP 404).")
+                if resp.status_code != 200:
+                    raise ProviderTransientError(
+                        f"Summarization failed (HTTP {resp.status_code}).")
+                content = await _collect_reply(resp, ticker)
         except (httpx.TimeoutException, httpx.TransportError) as e:
             raise ProviderTransientError(f"Ollama unreachable: {type(e).__name__}") from e
-        if resp.status_code == 404:
-            # Missing model and missing route both 404 here; both are
-            # configuration, not weather.
-            raise ProviderConfigError("Ollama has no such model or route (HTTP 404).")
-        if resp.status_code != 200:
-            raise ProviderTransientError(f"Summarization failed (HTTP {resp.status_code}).")
-        try:
-            content = resp.json()["message"]["content"]
-        except (ValueError, KeyError, TypeError) as e:
-            raise ProviderTransientError("Ollama sent an unreadable reply.") from e
         import json as _json
 
         try:
@@ -78,3 +86,44 @@ class OllamaProvider:
         except ValueError as e:
             raise ProviderTransientError("Ollama reply was not JSON.") from e
         return parse_summary(data, self.model)
+
+
+async def _collect_reply(resp: httpx.Response, ticker) -> str:
+    """Reassemble the assistant reply from Ollama's streamed chat response.
+
+    Streaming answers are NDJSON (one JSON object per line, ``done`` on the
+    last); a server honoring stream=false returns one JSON body — both work.
+    The running character count feeds [ticker] for UI progress."""
+    import json as _json
+
+    ctype = resp.headers.get("content-type", "")
+    if "ndjson" not in ctype and "event-stream" not in ctype:
+        body = await resp.aread()
+        try:
+            content = _json.loads(body)["message"]["content"]
+        except (ValueError, KeyError, TypeError) as e:
+            raise ProviderTransientError("Ollama sent an unreadable reply.") from e
+        ticker(len(content or ""))
+        return content or ""
+
+    parts: list[str] = []
+    total = 0
+    async for line in resp.aiter_lines():
+        line = line.strip()
+        if not line:
+            continue
+        try:
+            obj = _json.loads(line)
+        except ValueError:
+            continue
+        piece = (obj.get("message") or {}).get("content") or ""
+        if piece:
+            parts.append(piece)
+            total += len(piece)
+            ticker(total)
+        if obj.get("done"):
+            break
+    content = "".join(parts)
+    if not content:
+        raise ProviderTransientError("Ollama sent an unreadable reply.")
+    return content