Programmatic API

OpenExpertise is designed to be used both from the CLI and as a library. Every action the CLI takes is available as a composable TypeScript function or class. This page lists all public packages and explains when to reach for each one.

Workspace package map

Package	What it exports	When to import
@openexpertise/schema	parseExperienceYaml, validateExperienceSpec, all schema types	Parsing and validating experience.yaml without running anything
@openexpertise/core	runExperience, LLMClient, NodeDispatcher, DispatcherRegistry, StateStore, EventBus, RunContext	Building a custom runner, embedding OE in your own app
@openexpertise/node-kinds-tool	ToolDispatcher	Registering the tool node kind in a custom DispatcherRegistry
@openexpertise/node-kinds-agent	AgentDispatcher, AnthropicLLMClient	Registering the agent node kind; Anthropic-backed LLM client
@openexpertise/node-kinds-skill	SkillDispatcher	Registering the skill node kind
@openexpertise/node-kinds-dataset	DatasetDispatcher	Registering the dataset node kind
@openexpertise/node-kinds-experience	ExperienceDispatcher	Registering the experience node kind (sub-experience composition)
@openexpertise/node-kinds-cli-agent	CliAgentDispatcher	Registering the cli-agent node kind (Claude Code / Codex / Gemini)
@openexpertise/llm-openai	OpenAILLMClient	Using an OpenAI-compatible endpoint instead of Anthropic
@openexpertise/evolution	EvolutionAdvisor, EvolutionProposal	Generating YAML improvement proposals after a run
@openexpertise/authoring	UltraExpertise, AnalysisOutput, SynthesisOutput	LLM-assisted experience.yaml authoring
@openexpertise/mcp-server	MCP server entry point	Exposing OE as an MCP tool to Claude Code
@openexpertise/tui	TUI dashboard entry point	Embedding the terminal dashboard in another CLI
@openexpertise/cli	CLI entry point	Usually invoked via oe; rarely imported directly
@openexpertise/skill-experience-creator	Skill definition files	The bundled SKILL.md skill used by oe ultra

Quick start

import { runExperience } from '@openexpertise/core'
import { parseExperienceYaml } from '@openexpertise/schema'
import { DispatcherRegistry } from '@openexpertise/core'
import { ToolDispatcher } from '@openexpertise/node-kinds-tool'
import { AgentDispatcher, AnthropicLLMClient } from '@openexpertise/node-kinds-agent'
import { readFileSync } from 'node:fs'

const spec = parseExperienceYaml(readFileSync('experience.yaml', 'utf8'))

const client = new AnthropicLLMClient()
const dispatchers = new DispatcherRegistry()
dispatchers.register(new ToolDispatcher())
dispatchers.register(new AgentDispatcher({ client }))

const result = await runExperience({
  spec,
  experienceDir: process.cwd(),
  dispatchers,
})

console.log(result.status, result.finalState)

In this section

• runExperience — the top-level async entry point
• LLMClient — the provider-neutral LLM abstraction
• NodeDispatcher — interface every node kind implements
• EvolutionAdvisor — post-run improvement proposals
• UltraExpertise — LLM-driven authoring pipeline
• StateStore — SQLite blackboard wrapper
• EventBus — structured run events