CAESAR¶
Self-hosted homelab AI assistant. A central brain (Praetor) coordinates a pool of pluggable worker agents (Legion), talks to Home Assistant for device control, accepts voice input, drives a web dashboard, and runs on whatever LLM you point it at.
Pre-alpha. Don't deploy this on anything you care about yet.
Start here¶
- Quickstart in 10 minutes — the literal happy path from fresh box to talking to the brain.
- Architecture — the moving parts and how they fit.
- Roadmap — what we're building, in order.
- Glossary — Praetor, Legion, and every other term.
- Security model — trust boundaries and what the policy engine is for.
- Security review — known gaps and severity, for the cautious operator.
- Configuration — every environment variable.
- Run a worker on another box — cross-host Legion setup using NATS NKEY auth.
- Add your own tool — write a Legion worker that the brain can call: capability, policy entry, tests.
- Proactive CAESAR — schedule the brain to start runs on its own and push notifications to your phone via ntfy.sh.
- Memory — what CAESAR remembers about you between chats: how facts get extracted, how to inspect and correct them.
- Picking a model — Anthropic, OpenAI, or local Ollama, and how to mix-and-match per task.
- ADR index — every architecture decision so far.
License¶
CAESAR is licensed under PolyForm Noncommercial 1.0.0. Personal and noncommercial homelab use is welcome; commercial use requires a separate agreement. See README.md for the short version.
Contributing¶
See CONTRIBUTING.md. Every contributor signs the CLA the first time they open a PR.