Vector DB Launch (Python Native Server)

ChromaDB provides the vector database backend for semantic search. If configured for Option C (Native Server) in vector_profiles.json, the database must be running as a background HTTP service to be accessed by operations.py.

When You Need This

• RAG ingest fails with connection refused to 127.0.0.1:8110
• Semantic search hangs or fails to connect
• The user has explicitly selected Option 2 (Python Native Server) during vector-db-init

Pre-Flight Check

# Check if ChromaDB is already running
curl -sf http://127.0.0.1:8110/api/v1/heartbeat > /dev/null && echo "✅ ChromaDB running" || echo "❌ ChromaDB not running"

If it prints "✅ ChromaDB running", you're done. If not, proceed.

Launching the Server (Native Python)

The ChromaDB server runs as a background Python process.

It binds to the ${chroma_host}:${chroma_port} defined in your active profile inside .agent/learning/vector_profiles.json (defaults to 127.0.0.1:8110). Its data volume is mounted from the path defined by the profile's ${chroma_data_path}.

Step 1: Start the Service via CLI

Instruct the user to start the server as a background process using nohup or & so it does not block their terminal. Example:

chroma run --host 127.0.0.1 --port 8110 --path .vector_data &

Step 2: Verify Connection

After the user confirms the server is running, verify it via API:

curl -sf http://127.0.0.1:8110/api/v1/heartbeat

It should return a JSON response containing a timestamp {"nanosecond heartbeat": ...}.

---

Troubleshooting

Symptom	Fix
chroma: command not found	The user hasn't run the vector-db-init skill yet. Run it to pip install chromadb.
Port 8110 already in use	Another process (or zombie chroma process) is using the port. lsof -i :8110 to find and kill it.
Permission Denied for data directory	Ensure the user has write access to the .vector_data directory.

Alternative: In-Process Mode

If the user decides they do not want to run a background server, you can instruct them to set chroma_host to an empty string "" in their profile in .agent/learning/vector_profiles.json.

The operations.py library will automatically fallback to "Option A" (PersistentClient) and initialize the database locally inside the python process without needing this skill.