Guide
Local AI starter
Get open weights running locally — runners, licenses, outbound call hygiene, and agent trust. No tokens-to-buy.
9 min read
Intent
Run models on hardware you control so prompts are not a default cloud log. Technology practice — not an investment pitch.
1. Pick a runner
Start with a well-documented local runner (Ollama-class UX or llama.cpp). Prefer projects that publish release checksums.
2. Check the license
Open weights ≠ “do anything.” Read the model card. Prefer sources that publish hashes and provenance.
3. Audit outbound calls
Local UI can still phone home for telemetry, embeddings, or tool APIs. Watch network activity on first run.
4. Keep keys separate
If an agent or plugin can spend or sign, treat it like a hot wallet: small blast radius, clear permissions. See wallets and for agents.
5. Go deeper
- Pillar: Decentralized AI
- Note: Decentralized and local AI as infrastructure
Educational only. DYOR.