REDDIT · REDDIT// 4d agoINFRASTRUCTURE

OpenClaw users report llama.cpp requests getting cancelled mid-prompt

A Reddit user reports that OpenClaw cannot reliably chat against a local llama.cpp server when using Gemma 4 and Qwen3.5. The model endpoint responds with HTTP 200, but OpenClaw appears to treat the run as a network failure while llama.cpp logs show the task being cancelled around 31% through prompt processing. The user notes that direct calls to the llama.cpp endpoint and llama.cpp’s own web UI both work, which points to an integration issue rather than a broken model server.

// ANALYSIS

This reads like an OpenClaw-to-llama.cpp compatibility bug, not a model crash.

–The server is healthy enough to return `200`, so the failure likely sits in request lifecycle handling, timeout behavior, or streaming expectations on the client side.
–The cancellation happens while prompt ingestion is still in progress, which suggests OpenClaw may be aborting long first-token latency runs before generation starts.
–The config uses a large `contextWindow` and `reasoning: true`; either could push request size or behavior into a path OpenClaw does not handle well.
–The post is useful as a reproducible setup report, but it is still anecdotal until someone isolates whether the trigger is the OpenAI-completions adapter, chat template kwargs, or OpenClaw’s timeout logic.

// TAGS

openclawllamacppgemma4qwen35local-llmtroubleshootingapi-integrationself-hosting

DISCOVERED

4d ago

2026-04-08

PUBLISHED

4d ago

2026-04-08

RELEVANCE

8/ 10

AUTHOR

UnderstandingFew2968