How fabriziocondesballesteros/agency-brain works
Agency Brain is documented as an experimental foundation for persistent, tool-using AI agents—not as a conventional chat wrapper. Its intended workflow starts with a user request, passes through agent reasoning and orchestration, draws on memory, knowledge, tools, and deterministic rules, then uses a provider to produce a result or interact with integrations; session state supports continuation. The repository tree contains no runtime source files, so these are documented architectural components, not verified implementations. All component paths are null; relationships are based on the README’s conceptual diagrams and descriptions.
Components 14
Agent workflow
- Agent reasoning
- Workflow orchestration
- Session state
Memory and knowledge
- Persistent memory
- Knowledge retrieval
- Knowledge base
Agent capabilities
- Tool execution
- Business rules
- Structured workflows
Providers and integrations
- Model provider
- External integrations
Other components
- User
- External model
- External services
Connections 13
- User → Agent reasoning: submits request
README.md - Agent reasoning → Workflow orchestration: plans execution
README.md - Workflow orchestration → Session state: maintains state
README.md - Agent reasoning → Persistent memory: uses memory
README.md - Agent reasoning → Knowledge retrieval: retrieves context
README.md - Knowledge retrieval → Knowledge base: retrieves knowledge
README.md - Agent reasoning → Tool execution: chooses tools
README.md - Agent reasoning → Business rules: applies rules
README.md - Workflow orchestration → Structured workflows: executes workflows
README.md - Agent reasoning → Model provider: requests model
README.md - Model provider → External model: invokes model
README.md - Tool execution → External integrations: acts through
README.md - External integrations → External services: interacts with
README.md