Concepts
These pages explain how RantAIClaw works, not how to use it. Read them when you want to:
- understand the runtime well enough to extend it
- decide where to plug in new behavior
- reason about failure modes and security boundaries
| Page | What it covers |
|---|---|
| Architecture | The trait + factory pattern, subsystem map, blast-radius tiers |
| Profiles | The per-profile workspace under ~/.rantaiclaw/profiles/<name>/ |
| Agent Loop | How a single message flows from channel to tool to reply |
| Autonomy Levels | The 3-level runtime enum and the L1–L4 onboarding presets |
| Channels & Providers | The two big extension points and why they are separate |
| Tools & Skills | What an agent can do — and the difference between a tool and a skill |
| Memory | Backends, hybrid recall, and persistence |
| Gateway & Live Config | The HTTP control plane: pairing, webhooks, triggers |
| MCP Servers | Model Context Protocol supervisor (process-only today) |
| Security Model | Sandboxing, allowlists, secret store, audit logging |
If you are looking for a step-by-step “first agent” walkthrough, that lives in Getting Started. If you are looking for what to type, see Reference.
Last updated on