Platypus

Self-hosted AI Agents for your whole team — on your infrastructure, your models, around the clock.

Platypus is an open-source, full-stack application for building AI Agents that reason, use tools, and keep working when you aren't watching. You bring the models — hosted, proxied, or running on your own hardware — and Platypus gives you the Agents, the tools they call, the Triggers that run them, and the multi-tenant boundaries that keep one team's work out of another's.

🌐 Visit the website at platypus.chat.

📚 Full documentation lives at docs.platypus.chat.

!A Platypus Chat where an Agent edits a scheduled Trigger through tool calls

💡 Why Platypus

• Sovereignty. Run against local or in-house models (Ollama, vLLM, Qwen, …) so internal data never leaves your infrastructure.
• Always-on. Agents run on triggers and keep working server-side after you close the tab — not on a laptop that has to stay awake.
• Under one roof. Agents, Boards, Dashboards, Sandboxes, MCP, Triggers, and Memory in one platform instead of a stack of stitched-together services.
• Provider-agnostic. Local _or_ frontier models, chosen per Agent, so you control the cost/capability trade-off.

✨ Key Features

• 🤖 Agents, Skills & Sub-Agents: Build an Agent once — model, instructions, and tools — then give it reusable Skills it loads on demand and sub-agents it can delegate to.
• 🧰 Built-in Tool Sets: Agents move Kanban cards, update Dashboards, post Notifications, manage Triggers, read the web, and even build other Agents and Skills.
• 🔌 MCP Support: First-class Model Context Protocol support, so Agents connect securely to local and remote data sources.
• 📡 A2A Endpoints: Expose an Agent to outside agents and chat platforms over the open A2A protocol, with a token per client and every conversation kept as an ordinary Chat.
• 🏖️ Sandbox: Shell and filesystem access inside an isolated, per-workspace execution environment, with pluggable Docker and SSH reference backends.
• 🧠 Memory: Facts and preferences are extracted from your conversations in the background and injected into future chats, so Agents remember you over time.
• 📋 Boards & Dashboards: Drag-and-drop Kanban boards and widget-based dashboards, both readable and updatable by Agents through built-in tools.
• ⏰ Triggers: Run Agents on a cron schedule, as a one-off, or in response to Workspace events like a card landing on a Board.
• 🔔 Notifications & Webhooks: Agents post Notifications to the Workspace, and HMAC-signed webhooks push Board and Notification events to your own systems, with per-event filtering and automatic retries.
• 🏢 Multi-Tenancy: Organizations and Workspaces isolate data and keep one team's work out of another's.
• 📐 Blueprints: Package shared Agents, Skills, MCP servers, and Providers once, then stand up a new team's Workspace with them in one step.
• 🌐 Provider Agnostic: Powered by the Vercel AI SDK — OpenAI, Anthropic, Google, Bedrock, and OpenRouter, plus Ollama, vLLM, and any OpenAI-compatible endpoint.
• 🧩 Plugins: Add Tool sets, Sandbox backends, and web-search backends without forking, using the typed @platypuschat/plugin-sdk.
• ⚖️ MIT Licensed: Open source and free to use, on hardware you control.

See it all come together in the daily board digest worked example.

🚀 Quick Start (Docker)

curl -fsSL https://platypus.chat/install.sh | [email protected] bash

Needs Linux or macOS (Windows via WSL) and Docker with Compose v2. The installer pins the latest release, generates a secret and admin password, starts the stack, and prints where to sign in.

[!CAUTION]

The printed credentials are temporary. Change the password after your first login!

Installer settings and the manual, clone-based install are in Deploy with Docker Compose.

Sign-in is email and password; there is no SSO/OIDC/SAML.

For configuration, providers, sandbox infrastructure, and production deployment, see the Self-Hosting guide.

📚 Documentation

The docs site is the single source of truth for setup, concepts, and reference material:

• Getting Started — quick start, first run, and the default admin account.
• Administering — managing Workspaces and sharing resources across an Organization.
• Self-Hosting — Docker Compose, configuration & environment, providers & auth, and sandbox infrastructure.
• Concepts — the domain model: Organizations, Workspaces, Agents, Skills, MCP, Sandbox, and Memory.
• Building with Platypus — agents & sub-agents, skills, tool sets, MCP servers, A2A endpoints, triggers, boards, dashboards, notifications, and webhooks.
• Extending — writing plugins: Tool sets, Sandbox backends, and web-search backends.
• Reference — backend and frontend configuration reference.

🏗️ Architecture

Platypus is a monorepo managed by Turborepo:

• apps/frontend: A responsive web interface built with Next.js, ShadCN, and Tailwind. It uses the AI SDK for real-time streaming responses.
• apps/backend: A high-performance REST API built with Hono.js running on Node.js. It handles agent logic, tool execution, and database interactions.
• apps/docs / apps/website: The docs and marketing sites.
• packages/schemas: Shared Zod schemas used by both frontend and backend for end-to-end type safety.
• packages/plugin-sdk: The published plugin contract, with packages/example-plugin as a reference implementation.

🗺️ Roadmap

Curious where Platypus is headed — and where it isn't? See ROADMAP.md for the project vision and the themes we're working towards. If you're planning a substantial contribution, read it (especially the non-goals) and open an issue describing it first so the effort lands well.

🤝 Contributing

Contributions are welcome! Please read our Contributing Guide for details on local development, branch naming, commit conventions, and how to submit a pull request.

New here? Check for good first issues, or pick up a well-scoped item from the Roadmap — a new Sandbox backend is a great first plugin.

Platypus logo by Thiings.co