Build agent systems your whole team can govern.

Orchetree is the open-source platform for designing, governing, and deploying AI agent systems as versioned, role-controlled agent trees.

Join the waitlist Read the docs

Orchetree Cloud · coming soon

Apache-2.0 Self-hostable BYOK
Orchetree/Customer Support System
Export MCPRun tree
orchestrator
ORCHESTRATOR
researcher
SPECIALIST
writer
SPECIALIST
validator
SPECIALIST
router
ROUTER
-80%+
researcher
System
Business
Execution
Examples
Output
Co-pilot
Assistant
Assistant
Suggestion
Update system prompt

The platform

Give agent systems a structure the whole team can follow.

As agent systems grow, responsibilities can blur, prompts become harder to trace, and understanding how a result was produced takes more effort. Agent trees make roles, handoffs, prompts, and execution paths explicit, with a clear owner for every change and a trace for every run.

Multi-tree architecture

Compose agent systems from reusable specialist trees.

Structure complex agent systems as focused specialist trees with explicit roles and handoffs. Each tree can be composed, versioned, executed, and exported independently.

  • Orchestrator trees delegate to specialist trees through structured handoffs
  • Snapshot-based versioning supports review and rollback
  • Export portable bundles, expose trees through MCP, or integrate through backend APIs
Multi-tree architecture
Co-pilot

An AI assistant that understands your agent trees.

Built as an Orchetree specialist tree, Co-pilot analyzes your system and presents improvements as structured, reviewable action cards.

  • Suggests prompt block patches with structured diffs
  • Highlights inconsistencies between connected nodes
  • Explains how each suggestion may improve quality or safety
Co-pilot
Governance

Every prompt has an owner. Every change is tracked.

Keep business intent and execution details distinct. Role-based editing lets business teams own what the system does while engineers control how it runs.

  • Business roles edit goal-level prompts; execution roles control runtime behavior
  • Conflict detection keeps simultaneous edits from colliding
  • Audit trails capture prompt changes, runs, and exports
Governance
SDK

Build agent trees from code, without losing structure.

Define agent trees in Python or Node.js, work alongside your coding agent, and sync the result into the visual editor as a versioned, inspectable architecture.

  • Python decorators and typed Node.js function registration
  • Connect coding agents through the SDK, backend APIs, or admin MCP tools
  • Idempotent sync preserves stable identity and UI-managed prompt edits
SDK
Observability

See every node. Trace every run.

Step-through run inspection shows inputs, outputs, and timing for every node execution. Debug failures, spot latency, and verify governance inline.

  • Node-level timeline with expandable input/output views
  • Co-pilot analysis of failed or suspicious runs
  • Export run traces for offline analysis or compliance
Observability

Node capabilities

Six common capabilities for composing agent trees

Retrieval

Semantic search, RAG pipelines, and context-aware document lookup across any source.

Analysis

Structured data extraction, classification, and multi-step reasoning chains.

Synthesis

Long-form generation, summarization, and multi-source content assembly.

Validation

Output verification, fact-checking, and quality gate enforcement.

Planning

Goal decomposition, task scheduling, and adaptive multi-step execution.

Routing

Conditional branching, model selection, and dynamic handoff between specialists.

Bring your AI stack

Your keys. Your models. Your control.

Use built-in hosted providers or connect an OpenAI-compatible endpoint through BYOK. API keys are encrypted at rest and excluded from exported bundles.

Anthropic OpenAI Google Mistral Ollama Cloud OpenAI-compatible
BYOK · Your keys stay yours

Who it's for

Built for the whole team

For developers

  • Python decorators and typed Node.js function registration
  • Source-to-backend sync with stable resource identity
  • Inspectable planner and runner execution
  • Self-host anywhere; no vendor lock-in
  • Portable bundles, backend APIs, and MCP access

For business teams

  • Visual tree editor — no code required
  • Role-based editing: business vs. execution prompts
  • Co-pilot suggests improvements in plain language
  • Cloud waitlist for managed hosting and building assistants
  • Audit trail for every prompt change and run

Pricing

Self-host the complete platform. Cloud adds managed operation.

Self-hosted
Free / forever
  • Full editor and SDK
  • Unlimited self-hosted trees
  • Explicit planner and runner
  • Backend APIs and MCP endpoint
  • Portable bundle export
  • Self-host anywhere
  • Apache-2.0 license
View OSS repository
Cloud
Coming soon

Join the waitlist for availability updates

  • Everything in self-hosted
  • Hosted workspace and managed upgrades
  • Managed execution and operations
  • Persistent public MCP endpoints
  • Connect local coding agents through MCP
  • Premium Specialist Trees for faster building
Join waitlist

Ready to build your first agent tree?

Self-host the complete platform today, or join the Cloud waitlist for managed operation, persistent MCP access, and premium building assistants.

Join the waitlist Read the docs
Hero
Platform
Features
Capabilities
Integrations
Who it's for
Pricing
Get started