How Trinity Works

Deploy autonomous AI agents on your infrastructure in minutes. Agents collaborate, state persists to GitHub, you control everything.

What is Trinity?

Trinity is sovereign infrastructure for autonomous AI agents. Deploy agents (Claude Code or Gemini CLI) in isolated Docker containers, let them discover and collaborate with each other via MCP, persist all state to GitHub, and monitor everything through live execution streaming and a real-time dashboard. Go from local agent to production autonomous system in minutes.

Web terminal + SSH access to any agent
Isolated Docker container per agent
Agents discover and collaborate via MCP
State persists to GitHub - nothing lost
Live execution streaming + replay timeline
Your infrastructure, your roadmap

Core Capabilities

Trinity implements the infrastructure required for truly autonomous AI systems. Agents that plan, collaborate, remember, and operate continuously—all under your control.

Agents That Collaborate

Agents discover each other through MCP, delegate tasks, and share results automatically. Build workflows where specialists work together without custom orchestration code.

MCP Server: 21 tools for agent discovery, communication, and coordination
Permission Controls: Fine-grained control over which agents can talk to each other
Parallel Execution: Spawn tasks across multiple agents simultaneously
Agent Discovery: Agents find each other's capabilities programmatically

State That Never Dies

Every conversation, file, and decision syncs to GitHub. Delete a container? State survives. Need to roll back? One command. This is the holy grail of agent state management.

GitHub Sync: All agent state version-controlled automatically
SQLite Persistence: Conversations survive container restarts
Instant Rollback: Restore any agent to any previous state
Full History: Complete audit trail of every decision

Complete Observability

Watch agents work in real-time with live execution streaming. Replay timeline lets you scrub through history. Know exactly what happened, when, and why.

Live Execution Streaming: Watch agents work in real-time as they execute tasks
Replay Timeline: Scrub through agent history with zoom controls
Live Dashboard: Real-time view of agent activity and collaboration
Execution Logs: Full transcript of every agent action

Full Autonomy, Full Control

Agents run in isolated containers with complete autonomy within boundaries you define. Stop any runaway execution instantly. Centralized credentials. Resource limits. Autonomy toggles.

Execution Termination: Stop any runaway agent mid-execution instantly
Container Isolation: Each agent in its own Docker environment
Centralized Credentials: One place to manage all agent access
Resource Limits: Control memory (1-64GB) and CPU (1-16 cores) per agent

Architecture Overview

┌─────────────────────────────────────────────────────────────────┐
│                      TRINITY PLATFORM                           │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│  ┌───────────────────────────────────────────────────────────┐  │
│  │            ORCHESTRATOR AGENT (Claude Code)               │  │
│  │         Maintains master workplan, routes tasks            │  │
│  └─────────────────────────┬─────────────────────────────────┘  │
│                            │                                    │
│              ┌─────────────┼─────────────┬─────────────┐       │
│              │             │             │             │        │
│              ▼             ▼             ▼             ▼        │
│  ┌───────────────┐ ┌───────────────┐ ┌───────────────┐ ┌──────────────┐
│  │    WORKER     │ │    WORKER     │ │    WORKER     │ │   WORKER     │
│  │   Agent A     │ │   Agent B     │ │   Agent C     │ │   Agent D    │
│  │  (Content)    │ │  (Research)   │ │    (Code)     │ │  (Analysis)  │
│  │ Claude/Gemini │ │ Claude/Gemini │ │ Claude/Gemini │ │ Claude/Gemini│
│  │  YOLO Mode    │ │  YOLO Mode    │ │  YOLO Mode    │ │  YOLO Mode   │
│  └───────┬───────┘ └───────┬───────┘ └───────┬───────┘ └──────┬───────┘
│          │                 │                 │                 │
│          └─────────────────┴─────────────────┴─────────────────┘
│                                    │                            │
│                      ┌─────────────┴─────────────┐              │
│                      │        MCP Server         │              │
│                      │  (Agent Communication)    │              │
│                      └─────────────┬─────────────┘              │
│                                    │                            │
│  ┌─────────────────────────────────┴─────────────────────────┐  │
│  │                     INFRASTRUCTURE                         │  │
│  │  ┌────────────┐  ┌────────────┐  ┌────────────┐           │  │
│  │  │  Database  │  │   Docker   │  │    File    │           │  │
│  │  │ (Postgres) │  │ Containers │  │  Storage   │           │  │
│  │  └────────────┘  └────────────┘  └────────────┘           │  │
│  └───────────────────────────────────────────────────────────┘  │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘

How Agents Run

From creation to autonomous execution, here's how Trinity orchestrates Claude Code agents.

1

Agent Creation

  • Define agent template with CLAUDE.md
  • Specify permissions
  • Configure MCP servers and tools
2

Container Launch

  • Docker container spins up
  • Claude Code starts in YOLO mode
  • Workplan injected into context
3

Autonomous Execution

  • Agent reads workplan
  • Executes tasks via CLI
  • Delegates to sub-agents via MCP
4

Real-time Observability

  • Activity streams to dashboard
  • Custom metrics updated live
  • File changes tracked
5

Persistence & Recovery

  • Workplans persist across restarts
  • Chat history survives sessions
  • Failed tasks trigger replanning
6

Continuous Operation

  • 24/7 autonomous scheduling
  • Cron-based execution
  • Manual triggers available

What is YOLO Mode?

YOLO mode (You Only Live Once) is Claude Code's fully autonomous execution mode. When enabled, agents can operate without human confirmation for every action.

What agents can do

  • Execute any shell command
  • Write and modify files directly
  • Install packages and dependencies
  • Make network requests and API calls
  • Interact with external systems

Safety through infrastructure

  • Isolation: Each agent in its own container
  • Permissions: Control agent communication
  • Observability: See everything in real-time
  • Persistence: Full history of all actions
  • Sovereignty: Your infrastructure

Why this matters: Traditional AI assistants require human approval for every action. YOLO mode enables true autonomy—agents that can work for hours or days on complex tasks without human intervention. Trinity provides the infrastructure to run YOLO mode safely.

Platform Capabilities

Web Terminal + SSH

Direct access to any agent via browser or time-limited SSH

Multi-Runtime

Claude Code or Gemini CLI per agent—no model lock-in

Centralized Credentials

One place to manage all agent secrets and API keys

YAML Deployments

Deploy entire multi-agent systems with one manifest

Live Streaming + Replay

Watch agents work live, scrub through history

Scheduling + Autonomy

Cron-based execution with autonomy mode toggles

Deploy in 60 Seconds

Get Trinity running with three commands.

# Clone the repository
git clone https://github.com/abilityai/trinity
# Start Trinity
cd trinity && docker-compose up
# Access the dashboard
open http://localhost:3000
Web dashboard for agent management
Real-time activity monitoring
Workplan visualization
Custom metrics dashboard
File browser
Chat history viewer

What Can You Build?

Trinity enables autonomous AI systems that work for hours or days on complex tasks.

Autonomous Sales Intelligence

Deploy agents that monitor signals, qualify leads, and generate personalized outreach—running 24/7 without human intervention.

Competitive Response Systems

Agents track competitor moves, reason through implications, and draft response strategies autonomously.

Code Generation Pipelines

Multiple agents collaborate on complex codebases—one architects, one implements, one tests, one reviews.

Research & Analysis

Deep research agents that synthesize information from hundreds of sources into comprehensive reports.

Your Infrastructure. Your Control.

Security

Prompt injection is the #1 AI security risk of 2025. Sovereign deployment means you control the data perimeter.

Agility

Modify agent behavior tonight, not next quarter. No vendor roadmap dependencies.

Privacy

Your data never leaves your infrastructure. Air-gapped deployment available.

No Lock-in

Open source. Run anywhere. Migrate anytime.

Ready to Get Started?

Deploy autonomous agents on your infrastructure in minutes.