Source Available Autonomous Agent Framework

Build Coordinated
Autonomous Agent Teams

Six specialized autonomous agents working together like a real dev team. Genuine collaboration with observable decision-making, self-improvement, and defense-in-depth security.

View Docs GitHub →
Demo video coming soon

Built for Real Development

Not LLM wrappers — actual team coordination

🔗

Genuine Coordination

Lead routes tasks, prevents collisions, manages handoffs. Real role specialization.

👁️

Observable by Design

Every decision in Discord. Lurk, intervene, redirect anytime. Full transparency.

🔀

Multi-Organization

Switch clients with kernel isolation. Perfect for agencies and consultants.

Auto-Reload

System agent updates trigger team-wide reload. No manual restarts.

🔧

GitHub Native

Issues, PRs, Projects, CI/CD — full integration out of the box.

📡

Discord Native

Team communicates in channels you control. Real conversation threads.

Architecture Highlight

Self-Improving by Design

Autonomous agents can improve their own behavior — but safely. Every change goes through the same PR-based workflow you use for code.

  • System agent identifies improvement opportunities
  • Creates PR with proposed behavior changes
  • Human reviews and approves (or rejects)
  • Git-based rollback if anything goes wrong
1 System detects verbosity issue in responses
2 Creates PR: "Reduce response length in SOUL.md"
3 Human reviews diff, approves merge
4 All 6 autonomous agents reload with improvement
🌐

Network Firewall

Allowlist-only egress — autonomous agents can't exfiltrate data

📁

Filesystem Isolation

Kernel-enforced boundaries between organizations

🔑

Identity Tokens

Per-agent authentication with scoped permissions

📢

Channel Boundaries

Autonomous agents only access assigned channels

⚙️

Resource Limits

CPU, memory, and container hardening

Security Highlight

Zero-Trust Security

Defense-in-depth with 5 independent security layers. Each autonomous agent is isolated, authenticated, and resource-limited.

  • No data exfiltration possible (network allowlist)
  • Client data physically isolated at kernel level
  • Every action auditable in Discord
  • Runaway processes automatically killed

Get Started

Choose your setup path

Quick Start

Try It Out

Get running in 2 minutes. Agents work but can't self-improve.

$git clone https://github.com/sentai-ai/sentai.git
$cd sentai && sentai init
$sentai start

Ready to build?

BSL Licensed. Source available, free for development.

Read the Docs View on GitHub