Agent Orchestrator Framework (AOF)
A framework for building, running, and observing AI agents. Define agent blueprints, run them as managed sessions, and observe execution in real-time.
Design principles:
- Provider-agnostic — Executor layer abstracts the AI framework (Claude Code implemented, others pluggable)
- Minimal setup — Pre-built container images or local development scripts
- Observable — Dashboard for monitoring agent sessions and debugging
What is the Agent Orchestrator Framework?
Infrastructure for building AI agent systems. It provides:
- Agent Coordinator — Manages sessions, queues agent runs, tracks state
- Agent Runner — Executes agents via pluggable executors (Claude Code implemented)
- Context Store — Document storage and sharing between agents
- Dashboard — Real-time monitoring and session management
Use it to prototype agent workflows, build multi-agent systems, or as a backend for agent-powered applications.
See Architecture for detailed component interactions and terminology.
Quick Start
Option A: Container Images (Recommended)
Requires only Docker.
-
Get a Claude Code OAuth token:
claude setup-token
-
Create a .env file:
CLAUDE_CODE_OAUTH_TOKEN=<your-token>
-
Create the config directory:
mkdir -p config/agents
-
Start services:
docker compose -f docker-compose.quickstart.yml up
-
Open http://localhost:3000
Option B: Local Development
Requires Claude Code CLI, Docker, Python ≥3.11 + uv.
git clone <repo-url>
cd claude-agent-orchestrator
make start-all
Then start the Agent Runner in your project directory:
cd /path/to/your/project
/path/to/claude-agent-orchestrator/servers/agent-runner/agent-runner
Open http://localhost:3000
Stopping Services
# Containers
docker compose -f docker-compose.quickstart.yml down
# Local development
make stop-all
Repository Structure
├── README.md
├── Makefile # Build, run, deploy commands
├── docker-compose.yml # Full dev environment
├── docker-compose.quickstart.yml # Minimal setup with pre-built images
│
├── docs/ # Documentation
│ ├── ARCHITECTURE.md # System architecture overview
│ ├── GETTING_STARTED.md # Detailed setup guide
│ └── containers/ # Container image documentation
│
├── servers/
│ ├── agent-coordinator/ # Session management, agent registry, SSE events
│ ├── agent-runner/ # Polls coordinator, executes agent runs
│ │ ├── lib/agent_orchestrator_mcp/ # Embedded MCP server
│ │ └── executors/
│ │ └── claude-code/ # Claude Code executor (Agent SDK)
│ └── context-store/ # Document storage server
│
├── mcps/ # External MCP servers
│ ├── context-store/ # Document management MCP
│ ├── neo4j/ # Neo4j graph database MCP
│ ├── atlassian/ # Jira + Confluence MCP
│ └── ado/ # Azure DevOps MCP
│
├── apps/dashboard/ # Web UI (React + Vite)
│
├── plugins/ # Claude Code plugins
│ ├── orchestrator/ # ao-* orchestration commands
│ └── context-store/ # doc-* document commands
│
├── config/agents/ # Agent blueprints
├── scripts/ # Dev startup scripts
└── tests/ # Integration tests
MCP Servers
Optional MCP servers in mcps/ extend agent capabilities (Neo4j, Atlassian, Azure DevOps, Context Store).
The Agent Orchestrator MCP is embedded in the Agent Runner and automatically available to running agents.
See mcps/README.md for setup and configuration.
Example Agents
The config/agents/ directory contains example agent blueprints demonstrating various patterns: web research, Atlassian/ADO integration, knowledge management, browser automation, and more.
See config/agents/README.md for the full list.
Context Store
Document storage server for sharing context between agents. Supports tagging, querying, and parent-child document relationships.
Optional semantic search enables finding documents by meaning (requires Ollama + Elasticsearch).
See servers/context-store/README.md for setup and API documentation.
Service URLs
Testing