Troubleshooting

The model selector is empty

No LLM provider has an API key configured. Go to Settings → Preferences → General (or edit .env/~/.vectora/.env) and add at least one: GOOGLE_API_KEY, OPENAI_API_KEY, ANTHROPIC_API_KEY, COHERE_API_KEY, or configure a local Ollama endpoint.

RAG returns no results / embedding errors

RAG depends on COHERE_API_KEY (or a VoyageAI key configured as an alternative) to generate embeddings and rerank. Without it, indexing fails silently or vector search returns nothing. Check the key in Settings → Environment → Envs.

The agent can’t write files / run commands

The workspace is probably still untrusted. See First workspace — click “Trust this folder” to unlock writes and the terminal.

A connected MCP server’s tools don’t show up

Check the connector’s status in Settings → Environment → Plugins — a failed stdio/sse/http connection shows an error there. See MCP client.

vectora storage complete won’t connect to Postgres/Qdrant/Redis

Complete mode requires all three services to be reachable at the configured DSNs (POSTGRES_DSN, QDRANT_URL, REDIS_URL). If you don’t have your own infrastructure, run scons docker (from the monorepo root, if running from source) or use the vectora storage wizard to set up a managed provider (Supabase, Neon, Qdrant Cloud). See Storage: lite vs. complete.

Pro features (multi-user chat, complete storage) don’t unlock despite an active subscription

Confirm your VECTORA_TOKEN from the dashboard is configured and valid — license status is cached locally with a short TTL; force a revalidation by restarting the app or checking Settings → Administration → System.

command not found: vectora error (source install)

Run with uv run vectora ... instead of vectora ... directly — uv manages the virtual environment and entrypoint without a global install.

Where to report a bug

GitHub Issues (public) or the form at vectora.company/issues.