FastClaw / Run the model
Configuration
FastClaw does not have fastclaw.json: startup guidance relies on FASTCLAW_* environment variables, long-term configuration into the database, and can be modified through Dashboard or CLI.
No fastclaw.json
The README clearly says there is no fastclaw.json. Bootstrap settings only come from env; long-term configurations such as providers, channels, settings, defaults, etc. are all in DB and managed through Dashboard or CLI.
Why is it designed this way?
This makes FastClaw more like a server-side runtime: local development can be started with env, and production or multi-user scenarios leave variable configuration to the database and console.
Configuration scope
The merged config in the source code supports system, user, agent and other scopes. Agent can cover model, temperature, maxTokens, thinking, policy, sandbox, etc.; provider.<name>.* contains model supplier configuration.
| Scope | Common configurations | Description |
|---|---|---|
| System | plugins、skills、tools、objectstore、taskqueue、memory | Platform default behavior |
| Provider | provider.openai.apiKeyEnv、baseURL | Model vendor credentials and parameters |
| Agent | model、temperature、sandbox.enabled、promptMode | The running personality of a single Agent |
| Channel | token、account、agent binding | Binding accounts from different channels |
Clean keys after startup
cmd/fastclaw/main.go will call ScrubBootSecrets after the gateway is constructed to remove sensitive environment variables such as storage, objectstore, Redis, Boxlite, and E2B to reduce the probability of subsequent child processes accidentally reading the startup key.
- FASTCLAW_STORAGE_DSN
- Object storage access key
- Redis password
- Boxlite / E2B API key