Open Agent Harness Docs
Long-lived agent engine documentationOpen Agent Harness Docs
Section titled “Open Agent Harness Docs”Organize runtimes with Markdown and run many workspaces in parallel. OAH owns workspace lifecycle, agent execution loops, tool calls, queue coordination, and durable state; WebUI, TUI, and Desktop all connect through the same REST + SSE API.
TypeScript Engine
PostgreSQL + Redis
REST + SSE
Runtime Spec
pnpm install
export OPENAI_API_KEY=sk-...
pnpm dev:cli -- daemon start
pnpm dev:web
pnpm dev:cli -- tui
WebUI
TUI
Desktop
OAH API
Controller
Workers
Storage
Recommended Path
Section titled “Recommended Path”Start with runtime confidence, then move through configuration, architecture, and API integration.
Core Documentation Entrypoints
Section titled “Core Documentation Entrypoints”Organized around long-term maintenance tasks, not just a list of files.
Deploy and RunLocal development, embedded worker, split deployment, and Kubernetes paths.
Runtime ConfigBuild, publish, and maintain reusable runtime templates.
Workspace SpecSettings, agents, models, skills, actions, hooks, and MCP.
Engine InternalsLifecycle, context, execution backends, queue reliability, and audit events.
Kubernetes RolloutRollout checklist, production readiness, and operations runbook.
API and SchemasOpenAPI 3.1, server config schemas, and workspace config schemas.
System Shape
Section titled “System Shape”The docs should help maintainers answer: where should I change this, who does it affect, and how do I verify it?
Client SurfacesWebUI, TUI, and Desktop consume OAH-compatible APIs without owning engine state.
Control PlaneAPI, controller, queues, placement, and rebalance coordinate concurrent workspaces.
Execution PlaneEmbedded or sandbox workers execute runs and synchronize workspace state.
Storage PlanePostgreSQL is the source of truth, Redis coordinates queues, and object storage backs workspace state.