Zero-Config AI: Quickstart

Get unlimited AI working in 30 seconds. One command sets up local models, connects your Google account to the Gemini pool, and configures intelligent routing.

Step 1: Run the AI Doctor

nself doctor --ai

This command checks your system, installs Ollama if needed, pulls a recommended model for your hardware, and walks you through connecting a Google account to the Gemini pool.

Step 2: Verify

nself ai status

Expected output:

AI Stack Status
──────────────
Local (Ollama):   ✓ Running (gemma2:2b loaded, 14.2 tok/s)
Gemini Pool:      ✓ 1 account, 18/20 RPD remaining
Routing:          ✓ 11 task classes configured
Total capacity:   Unlimited local + 20 RPD free cloud

Step 3: Use It

AI is now available to any ɳSelf app. The routing engine handles everything automatically. For ɳClaw users, open the app and start chatting. AI requests route through the zero-config stack with no additional setup.

Optional: Add More Capacity

# Add another Google account (adds 20 RPD)
nself ai pool add

# Pull a larger local model
nself ai local pull llama3.1:8b

# Check what the router would pick for a task
nself ai route --task chat --dry-run

What Happens Behind the Scenes

1. nself doctor --ai detects available RAM and recommends a local model size.
2. Ollama is installed as a system service (systemd on Linux, launchd on macOS).
3. The recommended model is pulled and benchmarked.
4. If you connect a Google account, an OAuth flow creates a GCP project, enables the Gemini API, and generates a restricted API key. The key is AES-256 encrypted with a per-install DEK before storage.
5. Default routing rules are applied for all 11 task classes.
6. A health check confirms everything works.

Next Steps

• Local Models — Manage models, benchmarks, task assignment
• Gemini Pool — Add more accounts, monitor usage
• Routing — Customize tier priorities per task