Documentation navigation
Getting started Quickstart Installation Core concepts Product tour
Building trees Creating nodes Prompt blocks Connectors Roles and governance Skills Co-pilot Collaboration
Run and inspect Execution Runs and traces Versions and rollback
SDK Python SDK Node.js SDK
Integrate API MCP and bundles
Providers AI connections Hosted providers for testing
Deploy Self-hosting Orchetree Cloud Environment variables

AI connections

AI connections store the provider, model, endpoint, and credential used by nodes at runtime. Add and test them from Settings > AI Connections.

Supported connection families

  • Anthropic, OpenAI, Gemini, and Mistral
  • OpenAI-compatible hosted endpoints
  • Ollama and llama.cpp endpoints
  • Claude Code, Codex, Gemini, and GitHub Copilot command-line adapters

For a cloud deployment, use a network-accessible provider or endpoint. A model running only on a developer laptop is not reachable from a remote Orchetree instance.

Resolution order

For each node, the execution engine resolves the first configured connection in this order:

  1. node default
  2. tree default
  3. project default
  4. global default

Execution fails with a configuration error when none of these levels resolves to a connection.

Credentials

Provider keys are encrypted at rest. They are not returned as plain text by the application and are removed from exported bundles. Configure credentials again after importing a bundle into another environment.

Do not commit provider keys to source control or SDK definitions. Use the Settings interface, environment variables, or the destination platform’s secret manager.

Hosted testing

Free and trial allowances change frequently and are controlled by each provider. Review the current provider terms, rate limits, regional availability, and billing requirements before relying on an allowance.

See Hosted providers for the maintained setup links and usage cautions.