mofa-agent-swarm

Multi-agent swarm state is persisted in SQLite by storing agent identities, roles, concurrent task limits, and pending task queues in a robust rusqlite layer. This centralized state store enables reliable orchestration across the swarm.

Asynchronous task delegation with Tokio stores task metadata in the SQLite-backed state store for external executors to route and execute. The system is built on tokio and designed to integrate seamlessly with async runtimes.

Dynamic sub-agents like researchers, coders, and reviewers are spawned by recording their metadata and system prompts in the SQLite state store. A central orchestrator then creates and manages these specialized agents for multi-role workflows.

Task consistency when an agent shuts down is ensured through SQLite foreign keys that enforce cascading cleanup of tasks. This FK-enforced consistency automatically removes pending tasks associated with the terminated agent.

Prioritized tasks are fetched by consumers in rank order, with tasks ranked 1–10 and a default priority of 5. The priority-based task queue ensures higher-priority tasks are retrieved and executed before lower-priority ones.

An external orchestrator is required to manage the agent swarm, as the system provides a SQLite-backed state store rather than an execution engine. The orchestrator handles spawning, monitoring, and routing tasks to the surrounding async executors.