Repository navigation
Configuration
Agent B ships with no saved model connections. Open Settings → Add connection and choose a provider.
- API base:
https://api.openai.com/v1 - Model ID: the exact hosted model identifier
- Credential: an OpenAI API key
- API base:
https://api.anthropic.com/v1 - Model ID: the exact Claude model identifier
- Credential: an Anthropic API key
- Region: the AWS region used by the runtime endpoint
- Model ID: the exact model or inference-profile identifier
- Credential: a Bedrock bearer API key
Agent B uses the provider-neutral Converse API. Account and regional availability still determine which model IDs work.
Use this option for Ollama, LM Studio, vLLM, oMLX, Splash, llama.cpp, or another server exposing compatible /models and /chat/completions routes.
Typical API base:
http://127.0.0.1:8000/v1
Use the exact model ID returned by the server's /models endpoint. Enable Supports thinking control only when the server accepts chat_template_kwargs.enable_thinking.
- Each connection owns its credential.
- A credential is never copied to another connection.
- Public settings responses expose only a boolean indicating whether a key is set.
- Settings are written with owner-only permissions.
- Agent B redacts bearer credentials before saving transcript events.
The default API URL is:
http://127.0.0.1:8777
Keep this endpoint on loopback. Agent B accepts only loopback Agent A URLs.
Skills add focused methodology to a run without changing Burp's scope or safety controls. Select only what is relevant. A no-skill run is useful as an evaluation control.
DOUBLE_AGENT_URLAGENT_B_DATA_DIRAGENT_B_MODEL_URLAGENT_B_MODEL_API_KEYAGENT_B_MODELAGENT_B_MODEL_AUTH_FILEAWS_REGIONAWS_BEARER_TOKEN_BEDROCK
Avoid environment overrides for routine multi-provider use; saved connections provide stronger credential isolation.