Troubleshooting¶
Common issues and how to debug them.
Agent Not Responding¶
Check the daemon is running:
Check logs for errors:
Verify config is valid:
Telegram Bot Not Working¶
- Is Telegram enabled? Check
channels.telegram_enabledistrue - Is the token correct? Verify
TELEGRAM_TOKENin~/.vandelay/.env - Is the server publicly accessible? Telegram needs to reach your webhook URL
- Chat ID mismatch? If
telegram_chat_idis set, it must match your chat
Cron Jobs Not Firing¶
- Is the daemon running? Cron jobs only execute when the server is active
- Check the job list:
vandelay cron list - Timezone correct? Cron times are evaluated in the configured timezone
- Check logs: Look for scheduler-related entries in
vandelay daemon logs
High Token Usage¶
- Reduce history: Lower
num_history_runs(default: 2) ormax_tool_calls_from_history(default: 5) - Enable session summaries: Set
enable_session_summaries: trueto compress old sessions - Trim workspace files: Large
SOUL.mdorMEMORY.mdfiles increase every request's token count
Knowledge/RAG Issues¶
- "No embedder available": Install fastembed (
uv add fastembed) or set an explicit embedder provider - Empty results: Run
vandelay knowledge refresh --forceto rebuild the corpus - Slow refresh: Local fastembed depends on CPU speed. Consider using OpenAI embedder for faster processing.
Intel Mac (x86_64) — Knowledge Not Available¶
Neither chromadb nor lancedb ships pre-built wheels for macOS x86_64. This is an upstream limitation — onnxruntime (a dependency of both) dropped Intel Mac support in v1.19+.
Vandelay detects this automatically and disables knowledge during onboarding. You will see:
Your options:
| Option | Notes |
|---|---|
| Run without knowledge | Default — agent works fully, just no RAG |
| Build lancedb from source | Requires Rust toolchain — uv add lancedb will compile |
| Run a remote ChromaDB server | Point Vandelay at an external ChromaDB instance on a supported platform |
| Use an Apple Silicon or Linux machine | Full knowledge support out of the box |
To check your platform status at any time:
Memory Issues¶
- Agent not remembering: Check
update_memory_on_runis enabled (default: true) - Wrong channel memory: Verify all channels use the same
user_idin config - Memory too large: Use
vandelay memory clear --yesto reset agentic memory
Server Won't Start¶
- Port in use: Another process is using port 8000. Change with
VANDELAY_PORT=8001 - Missing API key: Check
~/.vandelay/.envhas your provider's API key - Import errors: Run
uv syncto install all dependencies
Getting Help¶
- Check the GitHub Issues
- Review the Configuration Reference for settings
- Enable debug logging for more verbose output