Skip to content

Open Agent Harness

A Headless Agent Runtime Kernel

Define agent logic in Markdown, switch agents by scenario, run many workspaces in parallel. You build the product UI. This is the backend runtime.

Get Started Architecture

What It Is

Open Agent Harness is a deployable backend runtime for agent products. It manages workspace lifecycles, agent execution loops, tool invocations, and state persistence. It does not ship a product UI — you bring your own frontend, and it runs the agents.

Core Capabilities

  • Parallel workspaces — PostgreSQL for persistence, Redis for queues and coordination. Many workspaces run concurrently.
  • Declarative agent config — Define agents in Markdown files with YAML frontmatter. Hot-reloaded.
  • Composable capabilities — agent / skill / action / tool / hook / context are configured independently per workspace.
  • Two workspace modes — project (full execution) and chat (read-only conversation).
  • REST + SSE API — Everything exposed under /api/v1. Frontend-agnostic.
  • Flexible deployment — Run unified locally or split into API + Worker for production.

Quick Start

pnpm install                                        # Install dependencies
pnpm infra:up                                       # Start PostgreSQL + Redis
pnpm dev:server -- --config ./server.example.yaml   # Start backend
pnpm dev:web                                        # Start debug console

After startup:

Full guide

Where to Go

  • Quick Start

    Install, launch, verify — up and running in 5 minutes

    Start

  • Architecture

    Layered design, core modules, request flow

    View

  • Workspace Config

    Agents, models, skills, actions, hooks

    Configure

  • Deploy and Run

    Local dev, split deployment, single workspace mode

    Deploy