Treat every workspace as a durable runtime unit.
OAH owns lifecycle, placement, state sync, and execution boundaries so UI surfaces do not need to re-implement agent orchestration.
Open Agent Harness Start Agent runtime for serious product teams
A long-term maintainable execution layer for AI products: run many workspaces in parallel, keep agent configuration reviewable, and expose one consistent API to WebUI, TUI, and Desktop clients.
Why OAH exists
Mature AI products need more than a chat box. They need deterministic workspace loading, runtime validation, replayable events, queue-aware execution, and docs that explain the contract between clients, control plane, workers, and storage.
OAH owns lifecycle, placement, state sync, and execution boundaries so UI surfaces do not need to re-implement agent orchestration.
Settings, prompts, models, tools, hooks, and MCP servers live as versioned workspace configuration instead of hidden app state.
Embedded workers keep development simple; split deployments add queues, controllers, and storage for larger multi-workspace runs.
Built for operation
Product teams can keep the user experience focused while OAH handles the runtime mechanics: controller decisions, worker execution, storage persistence, and streaming progress.
Explore engine shape$ pnpm dev:cli -- daemon start
workspace oah/demo loaded
runtime settings validated
controller queue online
worker pool ready
sse stream connected Site paths