openclaw - ✅(Solved) Fix [Bug]: LM Studio token stats not shown in /status (regression from 3.2 to 3.23) [2 pull requests, 1 participants]

SeananTQ · 2026-03-26T07:18:17Z

[openclaw] After upgrading OpenClaw from 2026.3.2 to 2026.3.23, /status shows Context: 0% and no token/cache info when using LM Studio backend. Works fine with… After upgrading OpenClaw from 2026.3.2 to 2026.3.23, `/status` shows Context: 0% and no token/cache info when using LM Studio backend. Works fine with Ollama and online models on the same version. Confirmed LM Studio API returns correct `usage` field via curl. # PR #55041: fix: use transcript usage as fallback for /status token display - Repository: openclaw/openclaw - Author: jjjojoj - State: closed | merged: True - Link: https://github.com/openclaw/openclaw/pull/55041 ## Description (problem / solution / changelog) Fixes #54995 ## Root Cause Both `/status` command and `session_status` tool call `buildStatusMessage` with `includeTranscriptUsage: false`. This disables the fallback path that reads token counts from the session transcript JSONL file. For custom providers like LM Studio, Ollama, or DashScope that use `api: openai-completions`, the agent meta store may not have usage data populated — so `/status` shows `Context: 0/131k (0%)`. ## Fix Set `includeTranscriptUsage: true` in both call sites: - `src/agents/tools/session-status-tool.ts` - `src/auto-reply/reply/commands-status.ts` This enables the existing fallback logic in `buildStatusMessage` which already has proper guards — it only uses transcript data when the meta store has zero/missing values, never overwriting valid data. ## Safety - `readUsageFromSessionLog` already handles missing files, parse errors, and missing sessionId gracefully - Merge guards: `totalTokens` only updated when zero or transcript value is larger; `inputTokens`/`outputTokens` only filled when missing - `/status` is a one-shot query, not a hot path — sync file read is acceptable - Confirmed working: online models (zai/glm-4.7) already had usage in meta store, so `includeTranscriptUsage: true` has no effect on them (guards prevent overwrite) ## Changed files - `CHANGELOG.md` (modified, +1/-0) - `src/agents/openclaw-tools.session-status.test.ts` (modified, +19/-0) - `src/agents/tools/session-status-tool.ts` (modified, +1/-0) - `src/auto-reply/reply/commands-status.test.ts` (modified, +83/-1) - `src/auto-reply/reply/commands-status.ts` (modified, +2/-1) --- # PR #55776: Agents: normalize WS usage aliases - Repository: openclaw/openclaw - Author: wangji0923 - State: closed | merged: True - Link: https://github.com/openclaw/openclaw/pull/55776 ## Description (problem / solution / changelog) ## Summary - Problem: the OpenAI-compatible WebSocket response parser still read usage from `input_tokens` / `output_tokens` only, so providers like DashScope that return `prompt_tokens` / `completion_tokens` recorded zero usage. - Why it matters: session transcripts and downstream usage displays lose token accounting for affected providers even though the provider returned valid usage data. - What changed: wired `buildAssistantMessageFromResponse` through the existing `normalizeUsage(...)` helper and added a focused regression test for `prompt_tokens` / `completion_tokens`. - What did NOT change (scope boundary): no provider-specific code, no UI/status rendering changes, and no changes outside the OpenAI-compatible response usage mapping path. ## Change Type (select all) - [x] Bug fix - [ ] Feature - [ ] Refactor - [ ] Docs - [ ] Security hardening - [ ] Chore/infra ## Scope (select all touched areas) - [ ] Gateway / orchestration - [x] Skills / tool execution - [ ] Auth / tokens - [ ] Memory / storage - [x] Integrations - [ ] API / contracts - [ ] UI / DX - [ ] CI/CD / infra ## Linked Issue/PR - Closes #54859 - Related #54995 ## User-visible / Behavior Changes - OpenAI-compatible providers that report usage as `prompt_tokens` / `completion_tokens` now record non-zero usage in assistant messages instead of defaulting those fields to zero. ## Security Impact (required) - New permissions/capabilities? (`Yes/No`) No - Secrets/tokens handling changed? (`Yes/No`) No - New/changed network calls? (`Yes/No`) No - Command/tool execution surface changed? (`Yes/No`) No - Data access scope changed? (`Yes/No`) No - If any `Yes`, explain risk + mitigation: ## Repro + Verification ### Environment - OS: Windows 11 - Runtime/container: source checkout - Model/provider: targeted unit coverage for OpenAI-compatible usage payloads (`prompt_tokens` / `completion_tokens`) - Integration/channel (if any): none - Relevant config (redacted): none ### Steps 1. Construct an OpenAI-compatible response object whose `usage` uses `prompt_tokens`, `completion_tokens`, and `total_tokens`. 2. Pass it through `buildAssistantMessageFromResponse(...)`. 3. Inspect the resulting assistant message usage fields. ### Expected - `usage.input` maps from `prompt_tokens` - `usage.output` maps from `completion_tokens` - `usage.totalTokens` remains populated ### Actual - Before this change, `usage.input` and `usage.output` defaulted to `0` because the parser only read `input_tokens`

openclaw2026-03-26 07:18:17

ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

GitHub issue URL

Helpful · Quick feedback

GitHub stats

openclaw/openclaw#54995•Fetched 2026-04-08 01:33:52

View on GitHub

Comments

Participants

Timeline

Reactions

Author

SeananTQ

Participants

SeananTQ

Timeline (top)

referenced ×6cross-referenced ×2labeled ×2closed ×1

After upgrading OpenClaw from 2026.3.2 to 2026.3.23, /status shows Context: 0% and no token/cache info when using LM Studio backend. Works fine with Ollama and online models on the same version. Confirmed LM Studio API returns correct usage field via curl.

Root Cause

RAW_BUFFERClick to expand / collapse

Bug type

Regression (worked before, now fails)

Summary

Steps to reproduce

Run LM Studio 0.4.7 with API server on (model: qwen/qwen3.5-35b-a3b, context length: 131k)
Configure OpenClaw 2026.3.23 to use LM Studio as model
Send any message, then type /status
Observe Context: 0/131k (0%) — no token/cache info displayed

Expected behavior

/status should show token count, context usage percentage, and cache info (e.g., Tokens: 18k/131k (14%))

Actual behavior

/status shows Context: 0/131k (0%) with no token or cache data

OpenClaw version

2026.3.23 (ccfeecb)

Operating system

Ubuntu 24.04 (VirtualBox VM on Windows 11 host)

Install method

npm global

Model

lmstudio/qwen/qwen3.5-35b-a3b

Provider / routing chain

OpenClaw -> LM Studio 0.4.7 (local API: http://192.168.x.x:1234/v1) -> qwen/qwen3.5-35b-a3b

Additional provider/model setup details

LM Studio API returns correct usage data: curl test shows "usage": {"prompt_tokens":13,"completion_tokens":10,"total_tokens":23}
Same OpenClaw 3.23 works with other backends:
- Online model (zai/glm-4.7): ✅ shows Context: 22k/205k (11%)
Regression confirmed:
- OpenClaw 2026.3.2 + LM Studio: ✅ works (see attached screenshot)
- OpenClaw 2026.3.23 + LM Studio: ❌ broken
Compaction still works (context management functional):
- After setting context length to 16k in config, conversation triggered auto-compaction (Compactions: 2), confirming only the display layer is broken, not actual context handling

Logs, screenshots, and evidence

Impact and severity

Affected: Users running OpenClaw with LM Studio backend on version 2026.3.23

Severity: Medium (affects monitoring visibility, does not block core functionality)

Frequency: Always (100% reproducible with LM Studio; 0% with online models)

Consequence: Cannot monitor token usage and context window occupancy via /status; must rely on compaction count as indirect indicator

Additional information

Regression: Last known good version 2026.3.2, first known bad version 2026.3.13 (confirmed 2026.3.13 and 2026.3.23 both affected)

Tested versions:

2026.3.2: ✅ Working
2026.3.13: ❌ Broken
2026.3.23: ❌ Broken

Workaround:

Use online models if token visibility is needed
Downgrade to 2026.3.2
Or rely on Compactions count to infer context pressure (functional despite display issue)

Additional context:

LM Studio API returns correct usage data (verified via curl)
Same OpenClaw versions work correctly with online models
Compaction triggers correctly in all versions (display-only issue)
Issue isolated to LM Studio adapter in OpenClaw

extent analysis

Fix Plan

To resolve the issue with OpenClaw 2026.3.23 not displaying token/cache info when using LM Studio backend, we need to update the LM Studio adapter in OpenClaw to correctly parse the usage field from the LM Studio API response.

Steps to Fix:

Update the LM Studio adapter in OpenClaw to handle the usage field correctly.
Modify the /status endpoint to display the token count, context usage percentage, and cache info based on the updated adapter.

Example Code:

// Updated LM Studio adapter
const fetchUsageData = async () => {
  const response = await fetch('http://192.168.x.x:1234/v1/usage');
  const data = await response.json();
  return data.usage;
};

// Modified /status endpoint
app.get('/status', async (req, res) => {
  const usageData = await fetchUsageData();
  const tokenCount = usageData.prompt_tokens + usageData.completion_tokens;
  const contextUsagePercentage = (tokenCount / 131k) * 100;
  const cacheInfo = `Tokens: ${tokenCount}/131k (${contextUsagePercentage}%)`;
  res.send(`Context: ${tokenCount}/131k (${contextUsagePercentage}%) - ${cacheInfo}`);
});

Verification

To verify that the fix worked, follow these steps:

Update the LM Studio adapter and /status endpoint with the modified code.
Restart the OpenClaw server.
Send a message and type /status to check if the token count, context usage percentage, and cache info are displayed correctly.

Extra Tips

Ensure that the LM Studio API is returning the correct usage field data.
Test the updated adapter and /status endpoint with different scenarios to ensure that the fix is working as expected.
Consider adding error handling and logging to the updated code to improve debugging and monitoring capabilities.

Vote matrix · Quick signals

Works

Did the solution work? Tap to confirm.

Easy Fix

Was it a quick fix?

Time Saver

Did it save you time?

Blocking

Was it severely blocking?

Common Issue

Are others likely hitting this too?

Flaky / Intermittent

Is it intermittent?

Verified / Reproducible

Can you reproduce it reliably?

FAQ

Expected behavior

/status should show token count, context usage percentage, and cache info (e.g., Tokens: 18k/131k (14%))

#api #output truncation #response parsing #generation error #database connection

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

openclaw - ✅(Solved) Fix [Bug]: LM Studio token stats not shown in /status (regression from 3.2 to 3.23) [2 pull requests, 1 participants]

Recommended Tools

GitHub issue graph ai analysis

Root Cause

Fix Action

Fix / Workaround

PR fix notes

PR #55041: fix: use transcript usage as fallback for /status token display

Description (problem / solution / changelog)

Root Cause

Fix

Safety

Changed files

PR #55776: Agents: normalize WS usage aliases

Description (problem / solution / changelog)

Summary

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

User-visible / Behavior Changes

Security Impact (required)

Repro + Verification

Environment

Steps

Expected

Actual

Evidence

Human Verification (required)

Review Conversations

Compatibility / Migration

Failure Recovery (if this breaks)

Risks and Mitigations

Changed files

Bug type

Summary

Steps to reproduce

Expected behavior

Actual behavior

OpenClaw version

Operating system

Install method

Model

Provider / routing chain

Additional provider/model setup details

Logs, screenshots, and evidence

Impact and severity

Additional information

extent analysis

Fix Plan

Steps to Fix:

Example Code:

Verification

Extra Tips

FAQ

Expected behavior

Still need to ship something?

RELATED_DISCOVERY

TRENDING