Issue evidence · Launch · openclaw/openclaw

agent exec with local Ollama times out after tool catalog; direct/gateway inference succeeds

GitHub issue: https://github.com/openclaw/openclaw/issues/168744

Not run

openclaw agent exec with a local qwen2.5-coder 1.5b hangs after cataloging 36 tools, while direct and gateway inference work.

What Badgr ran

Not run: it needs OpenClaw 2026.9.5 plus a local Ollama and a 3.8 GiB RAM, no-swap host to match the report, and the gateway setup was not built in this session.

What came back

  • The reporter's controls (direct /api/generate 9.4 s, gateway inference about 30 s) are in the issue.

The reporter's own question is whether a 1.5B model can handle tool calls at all; a larger model would not test that.

Checked 2026-10-11 in Badgr’s local development environment.

← All issue evidence