Skip to main content

Reference

This section is the alphabetical, exhaustive surface of every API the daemon exposes. It is the source of truth for fields, types, defaults, and runtime behaviour. The text under each entry assumes you already know why you are reaching for that primitive; for the mental model, see Concepts, and for guided learning paths see Tutorial.

Sections

SectionWhat it documents
ModulesEvery module shipped under (23 modules). One page per module, listing every @action, params model, return shape, and constraint spec.
RuntimeCross-cutting subsystems that aren't a single module: hooks, middleware, credentials vault, multimodal images, voice, configuration, tool chaining.
HTTP APIREST endpoints (/api/...) plus the Socket.IO event protocol and the DAP debug-adapter protocol.
CLIEvery digitorn ... sub-command, its flags, and a worked example.
Client SDKschat client, React preview SDK, Python testing SDK, web client spec.

Where things actually live

The runtime is an in-process Python framework. When you call a tool from the LLM, the path is:

  1. The LLM emits a tool call (native API or a parser-recoverable text format, see Agents - Native vs text-based).
  2. The agent loop dispatches the call through context_builder.execute_tool.
  3. The capabilities gate (App Configuration - tools.capabilities) verifies the action is allowed; high-risk actions pause for approval.
  4. The behavior engine (Behavior Engine) runs pre_tool_check then post-call post_tool_check.
  5. The module's @action-decorated method runs and returns a result.

Hooks (Hooks) fire around the call as configured.

How to navigate this section