IronClaw Architecture and Runtime
1. The workspace shape​
The README makes it clear that IronClaw's current story includes a dedicated Reborn runtime with its own binary and state root.
That already tells you something important:
the project is treating runtime boundaries and state layout as first-class design concerns
2. The runtime mental model​
At runtime, IronClaw generally works like this:
- the CLI resolves the active home and profile,
- configuration selects a model route,
- environment variables provide secrets,
- the runtime executes in the chosen policy and storage mode,
- outputs return through CLI surfaces like
run,repl, orserve.
3. Why the config home matters​
Because IronClaw is explicit about its config root, files, and seed behavior, the home directory is not a side detail. It is the operating boundary for:
config.toml,providers.json,- profiles,
- and runtime state.
That makes the architecture easier to reason about than systems that hide state across many locations.
4. Policy and deployment modes are part of the architecture​
The README shows that policy and storage are not just optional features. They shape real deployment modes, including production-style configurations with PostgreSQL and explicit policy sections.
That is why IronClaw feels more like an operator-friendly runtime than a hobby CLI.
5. What to read first in code​
Start with:
- the Reborn quick start in the README,
- config and provider docs,
- runtime and storage code,
- policy surfaces,
- skills or automation extensions after the core flow is clear.