Concaretti
High-orchestration agent swarm for complex task decomposition.
The Big Picture
What I Built
NestJS microservices monorepo. Each agent domain (research, coding, email, GitHub ops, file management) runs as an isolated NestJS module. BullMQ handles the task queue between agents. A Next.js 14 dashboard visualises the real-time reasoning graph. PostgreSQL stores agent memory and task history. Redis handles ephemeral state.
What Makes It Cool
- ✦17+ specialised agents
- ✦Deterministic .conca policy compliance
- ✦BullMQ task queue
- ✦Real-time reasoning dashboard
- ✦Private — school project with journal pending
Real-World Impact
Why This Matters
Current AI systems struggle with multi-step workflows, context preservation, and transparent reasoning. Concaretti solves this with 17+ specialized agents coordinating via deterministic .conca policies.
Role & Contributors
Behind The Scenes
How It Works
Agent actions are governed by .conca policy files — deterministic rules that define what each agent is permitted to do. No agent can escalate privileges beyond its declared scope. All external API keys are stored as encrypted environment variables, never exposed to the frontend.
Stack & Technologies
The Toolkit
The tools, frameworks, and engines picked specifically for this project:
Want to talk about this project or collaborate?
I'm always open to discussing new software ideas, apps, and opportunities.