A fully autonomous AI agent orchestration platform that transforms PRDs into runnable MVPs via multi-role collaboration (PM/Engineer/QA), featuring self-maintenance and Git-backed task persistence.
Overview#
Developed by Jordan Hubbard (FreeBSD co-founder), Loom is a full-stack AI software engineering platform built in Go. The core philosophy: "From a single thread of an idea, we weave complete software."
Core Capabilities#
Multi-Agent Orchestration#
- Built-in roles: Project Manager, Engineering Manager, Code Reviewer, QA Engineer, DevOps Engineer, Web Designer, Documentation Manager, Decision-maker
- Parallel execution with dependency management
- Autonomous decision-making within guardrails, with escalation support
Workflow Engine#
- Native Go goroutine ticker scheduling (Temporal dependency removed)
- Human-in-the-loop approval gates (CEO approval)
- Multi-step complex workflow support
Beads Work Item System#
- Git persistence: tasks recoverable after context compression
- Dependency tracking (blocking/blocked relationships)
- Four-tier priority queue: P0 (Critical) → P1 (High) → P2 (Normal) → P3 (Low)
Observability#
- Prometheus metrics collection
- Jaeger distributed tracing
- Grafana visualization dashboards
- SSE real-time event streaming
LLM Integration#
- TokenHub unified entry point: routing, failover, budgeting, billing
- Support for OpenAI-compatible API LLM providers
Enterprise Features#
- RBAC access control
- JWT/API Key authentication
- Multi-tenant isolation
Architecture#
Web UI → Control Plane API → Dispatcher → NATS JetStream → Agents Pool
↓
PostgreSQL
↓
Connectors Service
↓
Prometheus / Jaeger / Grafana
Key Components:
- Control Plane API: REST/gRPC core service
- Dispatcher: task distribution and scheduling
- NATS JetStream: messaging and event streaming
- Agents: stateless workers executing engineering tasks
- Beads: Git repository
.beadsdirectory for task state persistence
Quick Start#
Prerequisites: Docker & Docker Compose, LLM provider
git clone https://github.com/jordanhubbard/loom.git
cd loom
make start
Service Ports:
- Loom UI: http://localhost:8080
- TokenHub: http://localhost:8090
- Grafana: http://localhost:3000
Typical Use Cases#
| Scenario | Description |
|---|---|
| Rapid Project Launch | Auto-generate complete project skeleton from PRD |
| Autonomous Bug Fix | Detect → Investigate → Fix → Verify → PR Creation |
| Multi-Role Collaboration | PM/Engineering/QA/Design roles auto-coordinate |
| Continuous Maintenance | Self-maintenance capability, project evolves via own Agents |
Pending Confirmation#
- No official Hugging Face page
- No related academic papers
- Production deployment cases await community feedback