# LM Studio model passes a direct probe but the full agent still hangs **URL:** \ **Category:** Troubleshooting\ **Tags:** context, local-models, models, onboarding, windows\ **Created:** [September 14, 2026, 10:31pm UTC](https://community.openclaw.ai/t/lm-studio-model-passes-a-direct-probe-but-the-full-agent-still-hangs/84 "2026-09-14T22:31:15Z")\ **Posts on this page:** 8\ **Page:** 1
**Author:** ![ArchivedMember](https://avatars.discourse-cdn.com/v4/letter/a/e56c9b/32.png) [@ArchivedMember](https://community.openclaw.ai/u/ArchivedMember)\ **Post date:** [September 14, 2026, 10:31pm UTC](https://community.openclaw.ai/t/lm-studio-model-passes-a-direct-probe-but-the-full-agent-still-hangs/84/1 "2026-09-14T22:31:15Z")
> **Adapted community discussion:** Paraphrased and condensed from an anonymized public OpenClaw Discord support conversation from September 2026. Technical guidance was reviewed for OpenClaw 2026.9.3. I want to use a local vision-capable GGUF model with OpenClaw so I am not dependent on cloud-provider limits. Is that supported, and what needs to run between the model file and OpenClaw? ---
**Author:** ![OpenClawSupport](https://avatars.discourse-cdn.com/v4/letter/o/49beb7/32.png) [@OpenClawSupport](https://community.openclaw.ai/u/OpenClawSupport)\ **Post date:** [September 14, 2026, 10:31pm UTC](https://community.openclaw.ai/t/lm-studio-model-passes-a-direct-probe-but-the-full-agent-still-hangs/84/2 "2026-09-14T22:31:15Z")
Yes. OpenClaw does not read a GGUF file directly. A runner such as LM Studio must load it and expose a compatible local API. Local use avoids provider quotas and API charges, but hardware capacity, context size, output limits, request timeouts, and tool-call compatibility still apply. ---
**Author:** ![ArchivedMember](https://avatars.discourse-cdn.com/v4/letter/a/e56c9b/32.png) [@ArchivedMember](https://community.openclaw.ai/u/ArchivedMember)\ **Post date:** [September 14, 2026, 10:31pm UTC](https://community.openclaw.ai/t/lm-studio-model-passes-a-direct-probe-but-the-full-agent-still-hangs/84/3 "2026-09-14T22:31:15Z")
I previously removed OpenClaw and need to reinstall it on Windows before configuring LM Studio. I want to preserve any useful state and avoid selecting a cloud provider during setup. ---
**Author:** ![OpenClawSupport](https://avatars.discourse-cdn.com/v4/letter/o/49beb7/32.png) [@OpenClawSupport](https://community.openclaw.ai/u/OpenClawSupport)\ **Post date:** [September 14, 2026, 10:31pm UTC](https://community.openclaw.ai/t/lm-studio-model-passes-a-direct-probe-but-the-full-agent-still-hangs/84/4 "2026-09-14T22:31:15Z")
Use the supported Windows installer with onboarding skipped, then open a new terminal and verify both the OpenClaw CLI and Node. Do not reset the state directory merely because the CLI was uninstalled; review and preserve it unless you intentionally want a clean state. ---
**Author:** ![ArchivedMember](https://avatars.discourse-cdn.com/v4/letter/a/e56c9b/32.png) [@ArchivedMember](https://community.openclaw.ai/u/ArchivedMember)\ **Post date:** [September 14, 2026, 10:31pm UTC](https://community.openclaw.ai/t/lm-studio-model-passes-a-direct-probe-but-the-full-agent-still-hangs/84/5 "2026-09-14T22:31:15Z")
The installer completed, but PowerShell blocks the generated `.ps1` launcher under my execution policy. ---
**Author:** ![OpenClawSupport](https://avatars.discourse-cdn.com/v4/letter/o/49beb7/32.png) [@OpenClawSupport](https://community.openclaw.ai/u/OpenClawSupport)\ **Post date:** [September 14, 2026, 10:31pm UTC](https://community.openclaw.ai/t/lm-studio-model-passes-a-direct-probe-but-the-full-agent-still-hangs/84/6 "2026-09-14T22:31:15Z")
The installation itself succeeded. Use `openclaw.cmd` as the safe launcher instead of changing PowerShell execution policy. In LM Studio, load the model, start its local server, and test the model endpoint rather than the root URL. Confirm the exact model key before saving the local route in OpenClaw. ---
**Author:** ![ArchivedMember](https://avatars.discourse-cdn.com/v4/letter/a/e56c9b/32.png) [@ArchivedMember](https://community.openclaw.ai/u/ArchivedMember)\ **Post date:** [September 14, 2026, 10:31pm UTC](https://community.openclaw.ai/t/lm-studio-model-passes-a-direct-probe-but-the-full-agent-still-hangs/84/7 "2026-09-14T22:31:15Z")
The server is reachable even though a browser GET to its root URL is unsupported. Direct chat-completions requests return HTTP 200, and OpenClaw’s one-shot model probe returns `pong`. I enabled the lean local-model setting and increased the server context enough to clear the immediate overflow, but a full dashboard response still runs for minutes and hangs. ---
**Author:** ![OpenClawSupport](https://avatars.discourse-cdn.com/v4/letter/o/49beb7/32.png) [@OpenClawSupport](https://community.openclaw.ai/u/OpenClawSupport)\ **Post date:** [September 14, 2026, 10:31pm UTC](https://community.openclaw.ai/t/lm-studio-model-passes-a-direct-probe-but-the-full-agent-still-hangs/84/8 "2026-09-14T22:31:15Z")
That separates provider connectivity from full-agent behavior. The local server accepts the request, but the agent does not reach its end event before the run is aborted. Compare the dashboard run with an isolated local agent request: ```powershell openclaw.cmd agent --local --agent main --message "ping" ``` If that returns quickly, the full session or tool surface is the remaining issue. If it also hangs, test the model server’s streaming behavior next. Current reference: [local models](https://docs.openclaw.ai/gateway/local-models).