feat: ollama profile overlay and async WS parity

.env.example

@@ -24,6 +24,8 @@ POSTGRES_DB=synthora
 # --- LLM providers ---
 OPENAI_API_KEY=
 OPENAI_BASE_URL=                   # optional: OpenRouter / vLLM / LM Studio
+# When using docker compose --profile ollama with docker-compose.ollama.yml:
+# OLLAMA_BASE_URL=http://ollama:11434
 OLLAMA_BASE_URL=http://localhost:11434
 
 # --- Search ---

README.md

@@ -48,7 +48,13 @@ LangGraph Studio: `uv run langgraph dev` (uses `langgraph.json`).
 docker compose up -d
 ```
 
-Brings up: API (`:8000`), worker, web UI (`:3000`), Postgres, Redis, SearXNG. Optional local LLM: `docker compose --profile ollama up -d`.
+Brings up: API (`:8000`), worker, web UI (`:3000`), Postgres, Redis, SearXNG. Optional local LLM:
+
+```bash
+docker compose --profile ollama -f docker-compose.yml -f docker-compose.ollama.yml up -d
+```
+
+The overlay sets `OLLAMA_BASE_URL=http://ollama:11434` on API/worker. Default embeddings use `SYNTHORA_EMBEDDINGS=hash` (no Ollama required).
 
 Key environment variables (see `.env.example`): `SYNTHORA_DATABASE_URL`, `SYNTHORA_REDIS_URL`, `SYNTHORA_AUTH_MODE` (`none`|`session`), `OPENAI_API_KEY` / `OPENAI_BASE_URL`, `TAVILY_API_KEY`, `SEARXNG_URL`.
 

docker-compose.ollama.yml

@@ -0,0 +1,10 @@
+# Optional overlay when starting with `--profile ollama`.
+# Usage:
+#   docker compose --profile ollama -f docker-compose.yml -f docker-compose.ollama.yml up -d
+services:
+  api:
+    environment:
+      OLLAMA_BASE_URL: http://ollama:11434
+  worker:
+    environment:
+      OLLAMA_BASE_URL: http://ollama:11434

docs/feature-parity.md

@@ -140,6 +140,11 @@ Closed on ``feat/compose-embeddings-default``: compose defaults
 when profile disabled); sync ``SynthoraClient.iter_run_events``; embedding
 default tests; parity audit doc sync.
 
+Closed on ``feat/ollama-profile-and-async-ws``: ``docker-compose.ollama.yml``
+overlay wires ``OLLAMA_BASE_URL`` for the ollama profile; async
+``iter_run_events`` handles ``ConnectionClosed`` like sync; README/smoke
+comments updated.
+
 No known functional gaps remain beyond explicit non-goals below.
 
 Chat remains session-scoped ``fast_research`` with prior-report memory —

docs/parity-audit.md

@@ -1,7 +1,7 @@
 # Synthora parity audit checklist
 
 Living checklist against Open Deep Research, STORM/Co-STORM, and Local Deep
-Research. Last verified on `main` after PR #12 (`2537bd6`).
+Research. Last verified on `main` after PR #13 (`8e13595`) and ollama-profile overlay.
 
 ## Open Deep Research
 
@@ -69,6 +69,7 @@ Research. Last verified on `main` after PR #12 (`2537bd6`).
 | Item | Status |
 |---|---|
 | Compose default embeddings (`SYNTHORA_EMBEDDINGS=hash`) without Ollama profile | done (PR #13) |
+| Ollama profile overlay (`docker-compose.ollama.yml` sets service URL) | done (PR #14) |
 | Live compose smoke (`scripts/smoke.sh`: research, export, upload) | done |
 | Playwright UI e2e | API-mocked (CI speed); not a live-stack browser gate |
 

packages/sdk/src/synthora/sdk/async_client.py

@@ -159,14 +159,18 @@ async def wait_for_report(
     async def iter_run_events(self, run_id: str) -> AsyncIterator[dict]:
         """Stream live events from the run WebSocket."""
         import websockets
+        from websockets.exceptions import ConnectionClosed
 
         url = self.events_ws_url(run_id)
         headers = []
         if self.token:
             headers.append(("Authorization", f"Bearer {self.token}"))
         async with websockets.connect(url, additional_headers=headers) as ws:
             while True:
-                raw = await ws.recv()
+                try:
+                    raw = await ws.recv()
+                except ConnectionClosed:
+                    break
                 yield json.loads(raw)
 
     async def chat(self, message: str, *, session_id: Optional[str] = None) -> dict:

scripts/smoke.sh

@@ -4,8 +4,7 @@
 set -euo pipefail
 cd "$(dirname "$0")/.."
 
-# Compose defaults OLLAMA_BASE_URL to the optional ollama profile service; smoke
-# uses deterministic hash embeddings unless callers override these explicitly.
+# Compose defaults hash embeddings (no Ollama required). Override OPENAI/OLLAMA as needed.
 export OPENAI_API_KEY="${OPENAI_API_KEY:-}"
 export OLLAMA_BASE_URL="${OLLAMA_BASE_URL:-}"
 export SYNTHORA_EMBEDDINGS="${SYNTHORA_EMBEDDINGS:-hash}"

tests/test_sdk.py

@@ -209,6 +209,41 @@ def recv(self):
     assert events[-1]["type"] == "done"
 
 
+@pytest.mark.asyncio
+async def test_async_iter_run_events(monkeypatch):
+    from synthora.sdk.async_client import AsyncSynthoraClient
+    from websockets.exceptions import ConnectionClosedOK
+
+    payloads = [
+        {"type": "status", "message": "queued"},
+        {"type": "done", "message": "completed"},
+    ]
+
+    class _FakeWS:
+        def __init__(self):
+            self._queue = list(payloads)
+
+        async def __aenter__(self):
+            return self
+
+        async def __aexit__(self, *_args):
+            return False
+
+        async def recv(self):
+            if self._queue:
+                return json.dumps(self._queue.pop(0))
+            raise ConnectionClosedOK(None, None)
+
+    monkeypatch.setattr(
+        "websockets.connect",
+        lambda *_a, **_k: _FakeWS(),
+    )
+    client = AsyncSynthoraClient("http://localhost:8000")
+    events = [event async for event in client.iter_run_events("run-1")]
+    assert events[0]["type"] == "status"
+    assert events[-1]["type"] == "done"
+
+
 @pytest.mark.asyncio
 async def test_async_sdk_health_and_mcp(platform):
     from httpx import ASGITransport