Beta
AgentCore
A multi-agent workbench for building software
A local multi-agent system with specialised research, backend, frontend and monitoring agents sharing a persistent memory store across sessions.
The problem
A single general agent loses the thread on multi-day work. Context resets, decisions get re-litigated, and the same bug gets fixed twice.
What we built
Specialised agents with distinct roles write to a shared SQLite memory. Work survives session boundaries because decisions and findings are recorded outside any one agent's context window.
Capabilities
- Research, backend, frontend and monitor agents with distinct roles
- Shared SQLite memory persisting across sessions
- Project-scoped context so multiple products stay separated