ag3ntwerk

AI agent framework and orchestration toolkit. Build, coordinate, and deploy hierarchical multi-agent systems with 16 specialized agents managed by a central routing layer.

Part of the GozerAI ecosystem.

Features

• 16 specialized agents covering technology, finance, marketing, security, data, and more
• Hierarchical orchestration with central task routing and health monitoring
• REST API with JWT and API key authentication
• MCP server for tool integration with LLM clients
• Workflow engine with automation pipelines
• Learning system with agent improvement over time
• Web dashboard for monitoring and management

Agents

Agent	Domain	Description
Overwatch	Coordination	Task routing, health monitoring, drift detection
Forge	Technology	Development, architecture, DevOps
Keystone	Finance	Budgeting, cost analysis, financial reporting
Echo	Marketing	Content strategy, campaigns, brand
Sentinel	Information	Infrastructure governance, security policy
Blueprint	Product	Roadmap, feature planning, prioritization
Axiom	Research	Deep analysis, innovation, investigation
Index	Data	Data governance, knowledge management
Foundry	Engineering	Delivery, QA, CI/CD pipelines
Citadel	Security	Threat response, vulnerability management
Beacon	Customer	Experience, support, satisfaction
Vector	Revenue	Sales enablement, partnerships
Aegis	Risk	Risk assessment, BCP/DR
Accord	Compliance	Regulatory, audit, governance
Compass	Strategy	Strategic planning, market analysis

Quick Start

git clone https://github.com/GozerAI/ag3ntwerk.git
cd ag3ntwerk
pip install -e .

# With distributed support (Redis)
pip install -e ".[distributed]"

# Start the API server
uvicorn ag3ntwerk.api.app:app --host 0.0.0.0 --port 8000

Configuration

Variable	Default	Description
AGENTWERK_AUTH_REQUIRED	true	Enable/disable API authentication
AGENTWERK_API_KEY	—	API key for X-API-Key header auth
AGENTWERK_JWT_SECRET	—	JWT signing secret
AGENTWERK_LLM_PROVIDER	ollama	LLM backend (ollama, openai, gpt4all)
OLLAMA_HOST	http://localhost:11434	Ollama endpoint
OLLAMA_MODEL	—	Default model for agent reasoning

API Reference

Authentication: X-API-Key: <key> or Authorization: Bearer <jwt_token>. Auth is enabled by default.

Tasks & Workflows

Method	Endpoint	Description
POST	/api/v1/tasks	Submit a task for agent processing
GET	/api/v1/tasks/:id	Get task status and result
POST	/api/v1/workflows	Create a workflow
GET	/api/v1/workflows	List workflows
POST	/api/v1/workflows/:id/run	Execute a workflow

Agents & Fleet

Method	Endpoint	Description
GET	/api/v1/agents	List all agents
GET	/api/v1/agents/:name	Get agent details
GET	/api/v1/fleet	Fleet status overview
GET	/api/v1/fleet/health	Agent health checks

Dashboard & Metrics

Method	Endpoint	Description
GET	/api/v1/dashboard	Dashboard summary
GET	/api/v1/metrics	System metrics
GET	/api/v1/learning/progress	Agent learning progress

Conversations

Method	Endpoint	Description
POST	/api/v1/chat	Chat with an agent
GET	/api/v1/conversations	List conversations

MCP Server

ag3ntwerk includes an MCP (Model Context Protocol) server for integration with LLM clients:

# Start the MCP server
python -m ag3ntwerk.mcp.server

The MCP server exposes agent capabilities as tools, allowing LLM clients to dispatch tasks, run workflows, and query agent status.

Usage Example

from ag3ntwerk.core import AgentOrchestrator

orchestrator = AgentOrchestrator()

# Route a task to the best agent
result = await orchestrator.execute_task(
    "Analyze our Q1 revenue trends and identify growth opportunities"
)
print(result.agent)   # "Vector" or "Keystone"
print(result.output)

Docker

docker compose up -d

See docker-compose.yml for the full stack configuration.

License

AGPL-3.0 — see LICENSE for details. Commercial licenses available at gozerai.com.