IronClaw Configuration and Security
1. The core config mental model​
IronClaw configuration is intentionally explicit. The README shows a minimal route like:
[llm.default]
provider_id = "openai"
model = "gpt-5-mini"
api_key_env = "OPENAI_API_KEY"
That tells you the basic story immediately:
- routes are named and structured,
- secrets stay in the environment,
- config files describe the system instead of embedding credentials.
2. Secrets are env-only by design​
The upstream docs are especially clear on one point:
secret-shaped values do not belong inline in
config.tomlorproviders.json
That is one of the strongest trust signals in the project.
3. Profiles and deployment mode matter​
IronClaw's policy model means configuration is not only about choosing a model. It is also about deciding:
- local development vs production,
- storage backend,
- policy defaults,
- and what kinds of runtime behavior are acceptable.
4. Safe operating habits​
Use these defaults:
- isolated home path during evaluation,
- one provider first,
- environment-based secrets only,
- explicit profile selection,
- human review before broader deployment.
5. Practical rollout advice​
For teams, start with a local-dev profile and a small pilot. Only move toward remote storage or more hosted modes once the operating model is documented and owned.