claude-context: indexing succeeds but Milvus/Zilliz DB stays empty
Using zilliztech/claude-context with Claude Code, indexing reports success and the embedding model logs many calls, but the Milvus/Zilliz database stays empty. This happens with text-embedding-3-large, text-embedding-3-small, and local text-embedding-nomic-embed-text-v2-moe. The issue is still open as of November 2025. What causes this and how do I fix it?
Originally asked on GitHub: https://github.com/zilliztech/claude-context/issues/195
6 answers
Still open as of November 2025; no official fix was posted.
What the thread shows:
- One user fixed a similar issue by enabling debug logging (they launched their coder with
qwen --debug; Claude Code has a debug log capability too). The debug log exposed errors indicating the embeddings were broken, and the cause was a typo inOPENAI_BASE_URL. After correcting it, indexing worked. They suggest using a small codebase (a few files) so the log stays readable. - Another commenter says this silent failure happens when the OpenAI token is invalid or not properly set; the tool does not raise an error. Verify the token in
.env. - A debug log posted later shows the cloud-sync side: Zilliz Cloud collections
hybrid_code_chunks_c300ee9eandhybrid_code_chunks_5870516bexist but are empty, sync reportsFound 0 valid codebases in cloud, and then removes the local codebase from the snapshot (Indexed: 0). If your logs look like this, the local index is being discarded because the cloud collections hold no data. - For one user these checks were not enough:
.envwas correct, the API call worked withcurl, and SSL, token, and base URL were all verified, yet logs still showed a connection issue.
General advice: re-check
OPENAI_BASE_URLfor typos, wrong scheme, or trailing slashes, and confirm the model name in the URL matches the model id. Security note: keep API keys out of any debug output you paste into issues.Did this work for you?- One user fixed a similar issue by enabling debug logging (they launched their coder with
What a GitHub user tried on GitHub (Didn't work)
The reporter tried
text-embedding-3-large,text-embedding-3-small, and localtext-embedding-nomic-embed-text-v2-moe. Logs showed the embedding models being called many times, but the database stayed empty and Claude Code reported indexing success.Source: https://github.com/zilliztech/claude-context/issues/195 (opens external site)
Did this work for you?What @propell-ant tried on GitHub (Worked)
One user fixed a similar issue by finding a typo in
OPENAI_BASE_URL. They enabled debug logging (launched qwen coder withqwen --debug), which showed errors indicating the embeddings were broken; correcting the base URL resolved it.Did this work for you?What @sudoitir tried on GitHub (Didn't work)
A user with the same issue verified the
.envfile, confirmed the API call works viacurl, and checked SSL, token, and base URL. Logs still showed a connection problem, so those checks did not resolve it.Did this work for you?What @mahmoudnaif tried on GitHub (Outcome unclear)
Suggested checking the OpenAI token configuration, stating that an invalid or improperly set token causes this exact silent failure and that the tool should surface an error message instead.
Did this work for you?What @dan1els tried on GitHub (Outcome unclear)
Shared a debug log showing Zilliz Cloud collections
hybrid_code_chunks_c300ee9eandhybrid_code_chunks_5870516bare empty, sync finds 0 valid cloud codebases, and it removes the local codebase from the snapshot withIndexed: 0.Did this work for you?
Related questions
- claude-context: "Error validating collection creation: [object Object]" when indexing codebase
- How do I stop claude-context from applying .dockerignore when indexing?
- claude-context on Windows: 'Failed to open file' path has no slashes
- Can I use claude-context without an OpenAI key (OpenRouter, Mistral, or local LM Studio)?
- claude-context search_code: 'codebase is not indexed' after successful indexing?