fix(llm): use a JSON-serializable per-turn model timeout

[devin-ai-integration]

Summary

#802's per-turn timeout put an httpx.Timeout object into ModelSettings.extra_args. The Chat Completions and LiteLLM model paths build their tracing generation span from ModelSettings.to_json_dict() (evaluated even with tracing disabled), and pydantic cannot JSON-serialize httpx.Timeout:

model_settings.to_json_dict() → PydanticSerializationError:
Unable to serialize unknown type: <class 'openai.Timeout'>

This fails every model turn on those paths (only the Responses path, which doesn't serialize settings, was unaffected). Fix: pass the timeout as a plain float —

-return {"timeout": httpx.Timeout(timeout_s, connect=min(timeout_s, 30.0))}
+return {"timeout": timeout_s}

httpx-based clients apply a scalar as the read (inactivity) timeout, so the stalled-stream protection from #802 is preserved. Added a regression test asserting to_json_dict() succeeds with the timeout set.

Link to Devin session: https://app.devin.ai/sessions/dad023e379e942f287bcf6822463b7a4
Requested by: @0xallam

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[greptile-apps]

Greptile Summary

This PR makes the per-turn model timeout JSON-serializable and applies it to model requests. The main changes are:

• Adds a scalar timeout helper for model settings.
• Passes the configured timeout through scan and deduplication requests.
• Adds tests for serialization, propagation, disabled values, and retries.

Confidence Score: 4/5

The connection-timeout regression should be fixed before merging.

• The scalar value fixes ModelSettings serialization.
• Scan and deduplication requests receive the configured timeout.
• Failed connection attempts can now wait for the full model timeout instead of the former 30-second cap.

strix/config/models.py

Important Files Changed

Filename	Overview
strix/config/models.py	Adds the serializable timeout helper, but the scalar value removes the previous connection-timeout cap.
strix/core/inputs.py	Adds timeout propagation to shared model settings and preserves it across setting resolution.
strix/core/runner.py	Passes the configured LLM timeout into scan model settings.
strix/report/dedupe.py	Adds the configured timeout to deduplication model requests.
tests/test_inputs.py	Tests timeout propagation, omission, and preservation during reasoning resolution.
tests/test_model_retry.py	Adds coverage for retrying timeout and network errors.
tests/test_models.py	Tests timeout conversion and ModelSettings JSON serialization.

Prompt To Fix All With AI

Fix the following 1 code review issue. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 1
strix/config/models.py:28
**Connection Timeout Loses Its Cap**

A scalar HTTPX timeout applies to connection establishment as well as reads. With the default 300-second setting, an unreachable provider can now consume up to 300 seconds per connection attempt instead of the previous 30-second cap, and retries can multiply that delay.

_{Reviews (1): Last reviewed commit: "chore(llm): shorten timeout helper docst..." | Re-trigger Greptile}