本地 Ollama 挂起排查实战：`ollama launch openclaw` 无响应时的最小诊断路径

问题/场景：本地 Ollama 模型在 OpenClaw UI 长时间显示 typing 但无返回。前置条件：可访问 OpenClaw 配置、Ollama 服务与主机进程信息。实施步骤：1) 对比 `ollama run` 与 OpenClaw 内调用结果；2) 校验 provider `baseUrl` 与模型 ID；3) 复现实验并记录 CPU/日志；4) 停止 gateway 后确认推理进程是否残留。关键命令：`ollama run`、`openclaw gateway stop` 等。验证方法：简单输入（ping）可稳定在合理时间返回。风险与边界：该问题为回归报告，最终修复版本与根因需跟踪官方更新。来源归因：Issue #31577。

GITHUBDiscovered 2026-03-08Author veekurz

Prerequisites

Ollama daemon is running locally and target model can be loaded.
You can inspect OpenClaw config and restart gateway safely.

Steps

Run a baseline direct test: `ollama run <model>` and record first-token latency.
Launch OpenClaw path (`ollama launch openclaw`) and reproduce with `ping` to compare behavior.
Verify provider config (`models.providers.ollama.baseUrl` without `/v1`, model id mapping, reasoning flags).
If UI hangs, stop gateway and confirm whether Ollama inference process exits or leaks.
Keep reproducible logs/config snapshot and track upstream issue for patched releases.

Commands

ollama run qwen3:8b

ollama launch openclaw

openclaw gateway stop

openclaw gateway status

Verify

OpenClaw UI/TUI returns responses for local model prompts and inference processes exit cleanly after stop.

Caveats

Current data is from a bug report; fix confirmation requires upstream release notes（需验证）.
Do not kill random host processes while troubleshooting; isolate Ollama PID ownership first.

Source attribution

This tip is aggregated from community/public sources and preserved with attribution.

Open original source ↗

Visit original post