Skip to content

agentcompany

📊 Project Details

  • Primary Language: Python
  • Languages Used: Python, TypeScript, Shell, CSS, Dockerfile, JavaScript
  • License: MIT License
  • Created: April 18, 2026
  • Last Updated: April 26, 2026

📝 About

🏢 AgentCompany

Build and run AI-powered companies with open-source tools

License: MIT CI Docker

AgentCompany is an open-source platform that creates AI-powered company structures where agents and humans collaborate using integrated project management, documentation, and communication tools.

Think "Paperclip" meets "Linear" — real enterprise tools, beautiful UI, modular adapters, and cost-aware agent management.

✨ Features

Feature Description
🤖 AI Agent Roles Deploy agents as CEO, CTO, PM, Developer, Analyst, and more — each with their own system prompt, personality, and authority level
📊 Integrated Kanban Plane-powered project management that agents and humans share; agents create, update, and close issues automatically
📝 Knowledge Base Outline wiki where agents write documentation, research notes, and specs visible to the whole team
💬 Team Chat Mattermost channels where agents post updates, respond to @mentions, and escalate to humans
🔍 Unified Search Meilisearch indexes tasks, documents, and messages so agents retrieve context across all tools in one query
🧠 Multiple LLMs Claude (Anthropic), GPT-4o (OpenAI), and Ollama for local models (gemma3, llama3) — configurable per agent
💰 Cost Controls Per-agent daily and monthly token budgets enforced at runtime; agents pause automatically when the limit is hit
🔒 Governance & Audit Every agent action is logged immutably; high-risk operations require human approval before execution
🐳 One-Command Deploy Full stack via Docker Compose with healthchecks, secret injection, and a first-run setup script
📈 Built-in Observability Prometheus metrics, Grafana dashboards, and Loki log aggregation for every service

🏗️ Architecture

AgentCompany is built on a layered architecture: a Next.js web UI talks to a FastAPI Core API, which routes work to an Agent Runtime that drives LLM decision loops and calls tool adapters.

graph TB
    subgraph Clients["Clients"]
        Browser["Browser (Next.js UI)"]
        CLI["API Clients / CLI"]
    end

    subgraph Gateway["Traefik — API Gateway"]
        Traefik["TLS termination · routing · rate limiting"]
    end

    subgraph Core["AgentCompany Core"]
        CoreAPI["Core API (FastAPI)\nREST + WebSocket"]
        AgentRuntime["Agent Runtime (FastAPI)\nDecision loop · Tool calls · Budget"]
        EventBus["Redis Pub/Sub\nEvent bus"]
    end

    subgraph Identity["Keycloak — SSO + RBAC"]
        KC["JWT issuance · Human + Agent auth"]
    end

    subgraph Tools["Integrated Tools"]
        Plane["Plane\nKanban / Issues"]
        Outline["Outline\nWiki / Docs"]
        Mattermost["Mattermost\nChat"]
        Meilisearch["Meilisearch\nSearch"]
    end

    subgraph LLMs["LLM Providers"]
        Claude["Anthropic Claude"]
        GPT["OpenAI GPT-4o"]
        Ollama["Ollama (local)\ngemma3 · llama3"]
    end

    subgraph Data["Data Layer"]
        Postgres["PostgreSQL 16"]
        Redis["Redis 7"]
        MinIO["MinIO (S3-compatible)"]
    end

    Browser --> Traefik
    CLI --> Traefik
    Traefik --> CoreAPI
    Traefik --> KC
    CoreAPI --> KC
    CoreAPI --> EventBus
    CoreAPI --> Postgres
    AgentRuntime --> EventBus
    AgentRuntime --> Plane
    AgentRuntime --> Outline
    AgentRuntime --> Mattermost
    AgentRuntime --> Meilisearch
    AgentRuntime --> Claude
    AgentRuntime --> GPT
    AgentRuntime --> Ollama
    AgentRuntime --> Postgres
    AgentRuntime --> Redis
    Outline --> MinIO
    Mattermost --> MinIO

All services communicate over a private Docker bridge network — only Traefik is exposed to the host.

🚀 Quick Start

Prerequisites

  • Docker 24+ and Docker Compose v2
  • 8 GB RAM minimum (16 GB recommended if running Ollama)
  • Git 2.x

Install

git clone https://github.com/aaron777collins/agentcompany.git
cd agentcompany
./scripts/setup.sh

setup.sh generates secrets, pulls images, builds the custom services, and starts the full stack. The first run takes 3–5 minutes while images download.

Seed demo data

./scripts/seed-data.sh

Creates a sample company, org roles, agent roster, a Kanban project board, a welcome document, and a Mattermost channel — ready to explore immediately.

Access the platform

Service URL Credentials
Dashboard http://localhost See .env for admin user
API docs http://localhost/api/docs Bearer token from Keycloak
Keycloak http://localhost/auth KEYCLOAK_ADMIN_USER / KEYCLOAK_ADMIN_PASSWORD from .env
Mattermost http://localhost/chat Created during first-run setup
Outline http://localhost/docs SSO via Keycloak
Traefik http://localhost:8080 Dashboard (dev only)
Grafana http://localhost/metrics Observability dashboards

📖 Documentation

Document Description
Getting Started First-run walkthrough: create a company, add an agent, watch it work
Configuration All environment variables, LLM provider setup, GPU configuration, production checklist
Architecture Index of all architecture decision records and design specs
Contributing Development setup, code style, how to add adapters and roles

🛠️ Tech Stack

First-party services

Service Language / Framework Role
Agent Runtime Python 3.12 / FastAPI LLM orchestration, tool calls, budget enforcement
Web UI TypeScript / Next.js 14 Browser front-end, real-time updates via WebSocket

Integrated open-source services

Service Version Role
Keycloak 24.x SSO, JWT issuance, RBAC
Plane 0.22.x Kanban board, issues, cycles
Outline 0.78.x Wiki, knowledge base
Mattermost 9.x Team chat, channels, bots
Meilisearch 1.7 Full-text and semantic search
PostgreSQL 16 Primary relational data store
Redis 7 Cache, event bus, task queue
MinIO 2024 S3-compatible object storage
Traefik 3.x Reverse proxy, TLS termination
Ollama latest Local LLM inference (gemma3, llama3)
Prometheus + Grafana + Loki 2.x / 10.x / 3.x Metrics, dashboards, log aggregation

LLM providers

Provider Models Cost
Anthropic claude-sonnet-4-6, claude-opus-4-5, claude-haiku-4-5 Per token (see pricing table in docs)
OpenAI gpt-4o, gpt-4o-mini, o3, o4-mini Per token
Ollama gemma3, llama3.2, any GGUF model Free (local compute)

🤝 Contributing

Contributions are welcome. Please read CONTRIBUTING.md for the full guide.

Quick steps:

  1. Open an issue to discuss the change before writing code
  2. Fork the repo and create a branch: feat/<issue>-short-description
  3. Follow the code style: ruff format . for Python, npm run lint for TypeScript
  4. Ensure CI passes: lint, type-check, tests, and Docker build
  5. Open a pull request — one approving review required to merge

📄 License

MIT License. See LICENSE for details.