OpenWork is an open-source agent harness and workspace orchestrator developed by different-ai. It provides a structured execution environment for running, evaluating, and managing autonomous AI coding and workflow agents across isolated runtimes. In daily operations, OpenWork coordinates agent execution across workspace runtimes, modular skills directories, and containerized micro-sandboxes (including Rust-backed sandboxing). It allows developers to define execution contexts, assign skill sets, dispatch tasks to underlying language models, and monitor multi-step tool and command executions within controlled environments. Unlike standalone, single-binary terminal coding assistants, OpenWork functions as a platform harness. It prioritizes modular agent execution layers, safe sandboxed runtime environments, and integrated evaluation harnesses to test and benchmark agent reliability systematically. The project is distributed as an open-source monorepo with core packages and enterprise integration paths, allowing developers to self-host and customize runtime configurations.
Sandboxed Execution — Integrates micro-sandbox runtimes (including Rust-based environments) to safely execute shell commands and code. - Modular Skill System — Supports pluggable skill packages and agent capabilities defined within dedicated harness directories. - Workspace Runtime Management — Manages multi-environment workspace settings and agent states across sessions. - Evaluation Harness — Includes dedicated evaluation suites to benchmark and validate agent task completion and accuracy. - Monorepo Architecture — Structured codebase separating user-facing applications, shared libraries, and environment runtimes.
Use cases
Use Case: Sandboxed Agent Command Execution Developers running autonomous code generation tasks can dispatch commands into OpenWork's isolated micro-sandboxes. This prevents destructive filesystem modifications on host machines while granting the agent necessary execution privileges.
Use Case: Agent Capability Evaluation AI engineers can utilize the integrated evaluation pipeline to benchmark different model backends and agent skill sets against standardized task suites, ensuring consistent code generation performance.
Use Case: Custom Tool & Skill Integration Teams building domain-specific automation can define modular skills inside the harness, allowing the orchestrator to expose custom CLI tools and external APIs directly to the agent runtime.
Who it is for
OpenWork is built for software engineers, tool builders, and AI researchers who need an orchestrator and safe execution harness for autonomous agents. It is not intended for developers looking for a lightweight, single-command interactive terminal chat assistant.
Install guide
Installation Clone the repository and install dependencies using standard monorepo tooling (pnpm / Node.js runtime): bash git clone https://github.com/different-ai/openwork.git cd openwork pnpm install
First Run Configure your environment variables by copying the sample environment file and adding your API keys: bash cp .env.dev .env # Edit .env to configure model provider keys and workspace settings Build the core packages and runtimes: bash pnpm build
Example Session Launch the local agent workspace runtime in development mode: bash pnpm dev To run evaluation suites or trigger test tasks against the workspace sandbox: bash pnpm test:evals
Plandex — Plandex is an open-source terminal-based AI coding agent designed to handle complex, multi-file engineering tasks in large codebases. Created by…
OpenCode — OpenCode is an open-source AI coding agent developed by Anomaly that operates across terminal environments, desktop applications, and IDE extensions.…
OpenHands — OpenHands (formerly OpenDevin) is an open-source autonomous AI software development agent and orchestration platform. It provides an extensible runtime environment…
Subtask — Subtask is an open-source CLI agent harness and orchestration tool developed by zippoxer in Go. It is built to coordinate…
EchoCoding — EchoCoding is an open-source agent harness and audio-augmented developer tooling layer created by launsion-boop. Designed to run alongside AI coding…
pi-reflect — pi-reflect is an open-source agent harness and self-improvement extension for pi coding agents, developed by jo-inc. It inspects agent session…