Open Source Multi-Agent Framework
Run autonomous AI agent teams 24/7. Circuit breakers, sprint automation, task delegation — build your own one-person AI company.
Agent OS is a framework for orchestrating autonomous multi-agent teams. It provides the primitives to run AI agents that plan, execute, monitor, and self-correct — all without human intervention.
Inspired by real-world patterns from running Ralph's autonomous agent system (circuit breaker, exit detection, response analysis), Agent OS packages these battle-tested ideas into a reusable, extensible framework.
Built on top of OpenClaw (agent runtime) and Paperclip (task coordination).
We built a one-person AI company — @agentxagi — that runs 12+ agents around the clock, posting content, engaging with audiences, and growing a following on autopilot.
This is the framework that makes it possible.
Most "multi-agent" setups are demos. Agent OS is production code — scripts that have been running 24/7 for weeks, with circuit breakers, quality gates, and self-healing baked in.
┌─────────────────────────────────────────────────────────────────┐
│ AGENT OS │
│ │
│ ┌──────────────┐ ┌──────────────┐ ┌──────────────────┐ │
│ │ SPRINT │───▶│ TASK BOARD │───▶│ AGENTS │ │
│ │ CREATOR │ │ (Paperclip) │ │ (heartbeat) │ │
│ │ (4x / day) │ │ │ │ │ │
│ └──────────────┘ └──────────────┘ └────────┬─────────┘ │
│ │ │
│ ▼ │
│ ┌──────────────┐ ┌──────────────┐ ┌──────────────────┐ │
│ │ METRICS & │◀───│ QUALITY │◀───│ EXECUTION │ │
│ │ ANALYTICS │ │ GATES │ │ ENGINE │ │
│ └──────┬───────┘ └──────────────┘ └──────────────────┘ │
│ │ │
│ ▼ │
│ ┌──────────────────────────────────────────────────────────┐ │
│ │ CIRCUIT BREAKER │ │
│ │ ┌──────────┐ ┌─────────────┐ ┌────────────────────┐ │ │
│ │ │ CONTENT │ │ ENGAGEMENT │ │ FOLLOWING │ │ │
│ │ │ QUALITY │ │ SIGNALS │ │ GROWTH │ │ │
│ │ └──────────┘ └─────────────┘ └────────────────────┘ │ │
│ └──────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────┘
Three independent circuit breakers that can halt operations automatically:
- Content Quality — Stops posting if content isn't meeting engagement thresholds
- Engagement Signals — Monitors likes, retweets, replies for anomaly detection
- Following Growth — Halts growth tactics if following/follower ratio degrades
Each breaker has configurable thresholds, cooldown periods, and auto-recovery.
- Automatic sprint creation (4x per day)
- Task generation based on goals and historical performance
- Priority queuing and dependency management
- Paperclip-powered task board
- Agent assignment with skill matching
- 15-minute monitoring loops
- Status tracking and escalation
- Pre-execution content review
- Engagement prediction scoring
- Automated approval/rejection workflow
- Cron-based scheduling
- Heartbeat monitoring
- Self-healing on failure
- Graceful degradation
- Twitter/X (primary)
- LinkedIn (planned)
- Newsletter (planned)
- Extensible adapter system
git clone https://github.com/agentxagi/agent-os.git
cd agent-os
npm installEdit config/agents.yaml to define your agent team:
agents:
- name: content-creator
model: gpt-4o
schedule: "0 */6 * * *"
platforms: [twitter]
- name: engagement-bot
model: gpt-4o
schedule: "*/15 * * * *"
platforms: [twitter]# Start the sprint engine
npm run sprint
# Monitor agent status
npm run status
# Check circuit breaker state
npm run circuit-breaker| Module | Path | Description |
|---|---|---|
| Circuit Breaker | src/circuit-breaker/ |
Independent breakers for content, engagement, and growth |
| Sprint Creator | src/sprint/ |
Automated sprint generation and task planning |
| Coordinator | src/coordinator/ |
Paperclip integration, 15min monitoring loops |
| API Helper | src/api/ |
Paperclip REST API wrapper |
| Content Pipeline | src/content/ |
Content generation, review, and publishing pipeline |
| Templates | templates/ |
Ready-to-use agent configurations |
Sprint Creator (cron: 4x/day)
│
├─▶ Analyze current metrics
├─▶ Generate tasks based on goals
├─▶ Push tasks to Paperclip board
└─▶ Notify agents via heartbeat
Created → Assigned → In Progress → Review → Done
│ │
│ └─▶ Failed → Retry (3x) → Escalate
└─▶ Stale (>1h idle) → Reassign
CLOSED ──▶ OPEN (threshold breached)
│
└─▶ HALF-OPEN (cooldown expired)
│
├─▶ CLOSED (recovery successful)
└─▶ OPEN (still failing)
Agent OS powers @agentxagi, a fully autonomous AI growth agent running on Twitter/X.
| Metric | Value |
|---|---|
| Agents running | 12+ |
| Uptime | 24/7 |
| Sprints per day | 4 |
| Monitoring interval | 15 minutes |
| Circuit breakers | 3 (content, engagement, following) |
| Platforms | Twitter/X |
| Human intervention | Near-zero |
The system creates content, monitors engagement, adjusts strategy, and self-corrects — all autonomously. The circuit breaker ensures quality: if engagement drops, it stops posting and recalibrates before resuming.
- v0.2 — CLI tool (
agent-os init,agent-os deploy) - v0.3 — NPM package (
@agent-os/core) - v0.4 — Agent templates (newsletter, research, sales)
- v0.5 — Web dashboard for monitoring
- v1.0 — Hosted version (deploy your own AI company in 5 minutes)
We welcome contributions! Here's how to get started:
- Fork the repository
- Create a branch (
git checkout -b feature/my-feature) - Make changes and test thoroughly
- Commit with clear messages (
git commit -m "Add X for Y") - Push to your fork (
git push origin feature/my-feature) - Open a Pull Request
- Keep scripts portable (bash, no exotic dependencies)
- Add comments explaining non-obvious logic
- Update this README if adding new components
- Test circuit breaker logic carefully — it's the safety net
MIT License — use it, modify it, ship it. Build your own AI company.