Debug logs in Telegram

[jade roost]

Is it possible to get telegram to show live interactions with the backend lmstudio so that I can see the progress of queries being processed

[vernal perch]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[zinc wolf]

Yep — for Telegram you can enable OpenClaw’s preview/progress streaming. That gives you a live edited Telegram message while the turn is running, including tool/progress lines and/or partial answer text.

For “what is OpenClaw doing?” use:

{
  channels: {
    telegram: {
      streaming: {
        mode: "progress",
        preview: {
          toolProgress: true,
          commandText: "status"
        }
      }
    }

  }
}

If you want to watch the answer itself appear instead, use mode: "partial" or mode: "block".

Important distinction: Telegram can show OpenClaw progress and model output previews, but it won’t show LM Studio’s internal backend logs/token timings as chat messages by default. For raw model-stream debugging, use local raw stream logging instead; those logs can contain prompts/tool output/secrets, so I wouldn’t forward them to Telegram unless you really sanitize/control that channel.

Docs:
https://docs.openclaw.ai/concepts/progress-drafts
https://docs.openclaw.ai/concepts/streaming
https://docs.openclaw.ai/providers/lmstudio

If you try this and nothing shows live, first check your version with openclaw --version — Telegram tool-progress previews landed in recent builds, so version matters here.

[jade roost]

ok, but I'm not seeing the number of tokens per second

[zinc wolf]

Right — that part isn’t currently surfaced in Telegram.

OpenClaw’s Telegram streaming/progress can show:

• partial answer/progress draft
• tool/progress lines
• final token usage via /usage tokens or /usage full

…but tokens/sec from LM Studio is backend throughput telemetry, not one of the metrics OpenClaw puts into Telegram progress drafts today.

Best options right now:

1. Watch LM Studio’s own server/UI logs for tok/s.
2. Use OpenClaw /usage full if you just need final token counts/cost-ish accounting.
3. For debugging, enable raw stream/chunk logs locally and calculate throughput from timestamps — but don’t forward those logs to chat unless you’re careful, because they can include prompts/tool output/secrets.

Docs for the supported surfaces:
https://docs.openclaw.ai/concepts/usage-tracking
https://docs.openclaw.ai/help/debugging#raw-stream-logging-openclaw

So: possible to see live progress in Telegram, yes; live LM Studio tok/s in Telegram, not built-in at the moment.

[jade roost]

so there is no plugin or skill that could get that info

[zinc wolf]

Not a bundled one that I know of — I checked the installed skills too, and I don’t see an LM Studio/tok/s skill.

But the idea is viable. LM Studio’s newer REST API can return stats like:

"stats": {
  "tokens_per_second": 51.4,
  "time_to_first_token": 0.111,
  "generation_time": 0.954
}

The catch: OpenClaw’s normal LM Studio provider path is OpenAI-compatible chat completions, and Telegram progress drafts don’t currently expose those LM Studio-specific stats fields.

So the practical routes are:

1. Custom provider/plugin: best long-term fix. Capture LM Studio response stats and add them to usage/progress output.

2. Local proxy/sidecar between OpenClaw and LM Studio: measure stream timing / token count, then optionally send a Telegram/Discord status message. Hacky but doable.
3. Skill: possible only as an on-demand diagnostic, not a transparent “show tok/s for every backend request” solution. Skills don’t automatically intercept the model backend stream.

My recommendation: if you want this reliably for all queries, make it a feature request/plugin-provider thing, not a skill. GitHub issue is the right place: https://github.com/openclaw/openclaw/issues