Indexing and dictation issues
Recover from model downloads, hash fallback, reindex failures, Ollama embedding errors, and voice setup problems.
On this page
Semantic search is unavailable
Open Settings → Indexing. Confirm Enable codebase index is on. When off, codebase_search is removed from the tool catalog.
Read Index status:
- downloading or loading: wait for the active model phase.
- indexing: inspect file and chunk progress.
fallback_hash: dense embedding failed and local hash is serving as fallback.- error: preserve the message before changing settings.
Use Reindex workspace once after fixing the cause.
Local dense model download fails
Confirm Auto-download model, network access, disk space, and model-directory permissions. A partial model must pass required-file validation before it becomes ready.
Ollama embedder fails
Confirm Ollama is reachable and Ollama embedding model exists on that host. The default model is nomic-embed-text. A remote Ollama host receives the indexed text.
Dictation mic or transcription fails
Open Settings → Voice and confirm the selected Dictation engine:
- OpenAI needs a saved OpenAI key.
- OpenRouter needs a saved OpenRouter key.
- Local needs an installed Whisper model.
The engine is read on mic stop. Installation alone does not switch to Local.
Local Whisper fails
Check the model card for download, load, or error status. Try Unload before reloading. Use Delete … cache only when you intend to redownload required model files.
The supported local IDs are whisper-tiny.en and whisper-small.en. Local transcription is English.
Collect evidence
Record app version, platform, selected embedder or dictation engine, model ID, phase, progress text, and exact error. Do not delete indexes or caches before capturing the error unless the recovery step specifically requires it.