fix agent inference timeout on slow local models

2026-10-04 15:04:35 +03:00
parent 0f045b27d4
commit 4f6353cf88
3 changed files with 23 additions and 3 deletions
@@ -2207,7 +2207,9 @@ async def _execute_agent(
    read_only_tool_results: dict[str, tuple[Any, list[str]]] = {}
    remaining_tool_calls = MAX_AGENT_TOOL_CALLS - len(steps)
    for call_index in range(remaining_tool_calls + 1):
-        completion = await get_completion(current_payload)
+        # Agent reasoning and tool-selection can take several minutes on a small
+        # local model. Keep this aligned with the SSE client/request limit below.
+        completion = await get_completion(current_payload, timeout_seconds=600.0)
        final_message = completion["choices"][0].get("message", {})
        tool_calls = final_message.get("tool_calls") or []
        if not tool_calls: