Self-hosted AI orchestration with full data sovereignty

Run any LLM locally. Orchestrate agents. Keep your data on your hardware.

What is Dryade?

Dryade is an AI orchestration platform you run on your own hardware. Connect any LLM provider -- local models via vLLM or Ollama, or cloud APIs like OpenAI and Anthropic -- and orchestrate intelligent agents that actually do things. Your conversations, your data, and your models never leave your infrastructure. No telemetry, no cloud dependency, no vendor lock-in.

Quick Start

Get Dryade running in under a minute:

git clone https://github.com/DryadeAI/Dryade.git
cd Dryade
cp .env.example .env
docker compose up -d

Open http://localhost:3000 and start chatting.

By default, Dryade connects to a local Ollama instance. Edit .env to switch to any LLM provider. See the Configuration Guide for all options.

Manual setup (without Docker)

For developers who want to work directly with the Python codebase:

git clone https://github.com/DryadeAI/Dryade.git
cd Dryade
uv venv && source .venv/bin/activate
uv sync
cp .env.example .env
# Edit .env with your LLM provider settings

uvicorn core.api.main:app --host 0.0.0.0 --port 8080

For the frontend:

cd dryade-workbench
npm install
npm run dev

With local GPU inference (vLLM)

If you have an NVIDIA GPU, run a local model alongside Dryade:

docker compose --profile gpu up -d

This starts a vLLM instance serving Qwen3-8B. Update VLLM_MODEL in .env to change the model. See the Edge Hardware Guide for Jetson and DGX Spark configurations.

Features

Multi-Agent Orchestration

ReAct loop with three modes: Chat for conversations, Planner for structured reasoning, and Orchestrate for autonomous multi-step workflows.

MCP Server Integration

Connect external tools and services through the Model Context Protocol. Semantic and regex-based routing across all connected servers.

Knowledge Base / RAG

Ingest documents, build semantic search indexes, and give your agents access to your organization's knowledge.

Visual Workflow Builder

Drag-and-drop agent pipelines with ReactFlow. Design, connect, and execute complex multi-agent workflows visually.

Any LLM Provider

Works with Ollama, vLLM, OpenAI, Anthropic, Google, and any OpenAI-compatible API. Switch providers without changing your workflows.

Edge Hardware Native

Runs natively on NVIDIA Jetson and DGX Spark with local models. Built for environments where data cannot leave the device.

Plugin Ecosystem

Extend Dryade with marketplace plugins for monitoring, security, compliance, and vertical industry workflows. Create your own plugins and sell them on the Dryade marketplace.

Full API

REST and WebSocket APIs with real-time streaming. OpenAPI-documented endpoints for integration with your existing tools and pipelines. See the API Reference.

Architecture

User <-> Workbench UI (React/TypeScript)
              |
         FastAPI Backend
              |
     DryadeOrchestrator (ReAct loop)
        / | \
   Chat Planner Orchestrate
              |
   HierarchicalToolRouter
     (semantic + regex matching)
        / \
  MCP Servers Plugins
        |
  LLM Providers
  (vLLM / Ollama / OpenAI / Anthropic / Google)

Framework Adapters

Dryade connects to agent frameworks through a unified adapter pattern:

Framework	Description
MCP	Model Context Protocol -- native integration with MCP servers
CrewAI	Multi-agent crew orchestration
ADK	Google Agent Development Kit integration
LangChain	LangChain tool and chain execution
A2A	Agent-to-Agent protocol for inter-agent communication

Deployment Options

Method	Best For	Guide
Docker Compose	Most users, production	Deployment Guide
Manual (Python + uv)	Development, contributors	Getting Started
Edge Hardware	Jetson, DGX Spark, local AI	Edge Guide

Comparison with alternatives

Feature	Dryade	Dify	n8n	Langflow
Self-hosted	Yes	Yes	Yes	Yes
Data sovereignty (zero telemetry)	Yes	Partial	Partial	Yes
Edge hardware support (Jetson, DGX)	Yes	No	No	No
MCP server integration	Yes	No	No	No
Multi-framework adapters (CrewAI, ADK, A2A)	Yes	No	No	No
Visual workflow builder	Yes	Yes	Yes	Yes
Plugin marketplace	Yes	No	No	No
RAG / Knowledge base	Yes	Yes	No	Yes
Local LLM support (vLLM, Ollama)	Yes	Yes	Partial	Yes
REST + WebSocket API	Yes	Yes	Yes	Yes
Real-time streaming	Yes	Yes	Yes	Yes

Community

Discord -- discord.gg/bvCPwqmu for questions, discussions, and support
GitHub Discussions -- Q&A, Ideas, Show & Tell
Contributing -- Read the Contribution Guide to get started
Examples -- Quickstart projects to get running in 5 minutes
Marketplace -- Create your own plugins and sell them on the Dryade marketplace
Documentation -- dryade.ai/docs for guides, API reference, and tutorials

Star History

Contributors

Roadmap

A2A Server Endpoint -- Expose Dryade agents for discovery by external orchestrators
Enhanced Plugin SDK -- Improved developer experience for building and distributing plugins
Kubernetes Deployment Guide -- Helm charts and production-grade K8s deployment
More Framework Adapters -- Expanding the adapter pattern to additional agent frameworks
See the GitHub Milestones for detailed progress.

License

Dryade is licensed under the Dryade Source Use License (DSUL). Enterprise features in core/ee/ are under separate terms.

Contact

Commercial licensing: [email protected]
Security issues: [email protected] (see Security Policy)
General inquiries: dryade.ai