Summary
While testing @memtensor/memos-local-openclaw-plugin on Windows, I hit a Gemini Viewer/Test-Connection bug in the installed package that I could reproduce locally, but I could not find an equivalent code path in the current apps/memos-local-openclaw source tree.
This suggests the published package and the current GitHub source may be out of sync, or that the relevant Viewer logic is generated / bundled differently from what is currently checked in.
What I observed in the installed package
In the locally installed plugin package under ~/.openclaw/extensions/memos-local-openclaw-plugin, the Viewer-side Gemini model test logic behaved incorrectly:
- Gemini embedding test used the wrong API shape/path:
- used
v1/...:embedContent
- expected behavior was
v1beta/...:batchEmbedContents
- It also ignored the user-provided endpoint value during testing.
This produced errors like:
{
"error": {
"code": 404,
"message": "models/text-embedding-004 is not found for API version v1beta, or is not supported for embedContent. Call ListModels to see the list of available models and their supported methods.",
"status": "NOT_FOUND"
}
}
Additional context
While debugging, I also confirmed that:
text-embedding-004 is no longer a good default for Gemini embedding in this environment
gemini-embedding-001 works against the official Gemini API
I already opened a separate PR for the repo-side fixes I could clearly trace and validate in source:
That PR covers:
- Windows path handling /
better-sqlite3 resolution
- Gemini embedding default model / endpoint normalization in source
Why this issue is separate
I did not include the Viewer Gemini test fix in the PR above because I could not confidently find the same buggy code path in the current repo source tree. I did not want to submit a speculative patch against the wrong file.
Request
Could you please check whether:
- the npm-published package for
@memtensor/memos-local-openclaw-plugin is ahead/behind the current repo source,
- the Viewer Gemini test logic is bundled from a different source path,
- or the publish pipeline is shipping stale built assets?
If helpful, I can provide the exact installed file path and the local patch I used to make the Viewer Gemini test behave correctly.
Summary
While testing
@memtensor/memos-local-openclaw-pluginon Windows, I hit a Gemini Viewer/Test-Connection bug in the installed package that I could reproduce locally, but I could not find an equivalent code path in the currentapps/memos-local-openclawsource tree.This suggests the published package and the current GitHub source may be out of sync, or that the relevant Viewer logic is generated / bundled differently from what is currently checked in.
What I observed in the installed package
In the locally installed plugin package under
~/.openclaw/extensions/memos-local-openclaw-plugin, the Viewer-side Gemini model test logic behaved incorrectly:v1/...:embedContentv1beta/...:batchEmbedContentsThis produced errors like:
{ "error": { "code": 404, "message": "models/text-embedding-004 is not found for API version v1beta, or is not supported for embedContent. Call ListModels to see the list of available models and their supported methods.", "status": "NOT_FOUND" } }Additional context
While debugging, I also confirmed that:
text-embedding-004is no longer a good default for Gemini embedding in this environmentgemini-embedding-001works against the official Gemini APII already opened a separate PR for the repo-side fixes I could clearly trace and validate in source:
That PR covers:
better-sqlite3resolutionWhy this issue is separate
I did not include the Viewer Gemini test fix in the PR above because I could not confidently find the same buggy code path in the current repo source tree. I did not want to submit a speculative patch against the wrong file.
Request
Could you please check whether:
@memtensor/memos-local-openclaw-pluginis ahead/behind the current repo source,If helpful, I can provide the exact installed file path and the local patch I used to make the Viewer Gemini test behave correctly.