OpenWork
openwork.software- Category
- AI Agents
- Rank
- No. 213Tools index
- Pricing
- Open Source
- Type
- APP
- Use case
- Workflow Automation · Productivity & Collaboration
- Interfaces
- Desktop
- Builder
- @different-ai
- GitHub
- 23.7k stars
- Latest release
- v0.18.52
- Date
About
An open source desktop application that provides access to 50+ large language models with the ability to bring your own API keys, automate browser tasks, and share AI workflows with your team. It serves as an alternative to Claude's Cowork feature with enhanced customization and team collaboration capabilities.
What it does
OpenWork turns reusable AI capabilities into a shared layer across desktop apps, coding agents, teammates, and machines. A compatible agent connects to its remote protocol server, searches assigned capabilities, then runs them. Organizations can publish skills and plugins, configure connections, allocate model access, and apply desktop policies through a central control plane.
Why it's ranked here
The strongest reason to consider OpenWork is portability. Teams can define capabilities once and expose them through one connection to several compatible agents. Its desktop workspace remains optional for that workflow. Central assignment, access controls, provider provisioning, and per-user connections make the design notably more useful for teams than a desktop-only assistant.
What's good
The interface stays intentionally small: agents search for available capabilities, then execute one. That reduces client integration work while letting administrators change assignments centrally. Skills, plugins, connected services, and remote protocol connections can travel together. The repository also includes explicit schemas, transport checks, automation components, headless thread support, evaluations, end-to-end tests, and isolated development profiles.
Tradeoffs
The broad architecture carries operational weight. Local development includes Electron, web services, MySQL, Redis, authentication, inference, and multiple ports. Organization deployment therefore appears more involved than installing a single desktop client. The separate interface-control server also requires the desktop app and its local bridge to be running. Remote access opens a browser for sign-in and organization selection.
How to use it well
Use OpenWork when several people or agent clients should share governed skills, plugins, services, and model access. Start with the remote connection inside an existing compatible agent, then add the desktop workspace when a dedicated interface helps. Adopt the organization control plane when assignments and policies justify its infrastructure. Do not treat the local interface-control server as a headless substitute for the desktop app.
Technical notes+
The root package.json defines a private pnpm 11 workspace with TypeScript tooling, Electron development commands, Den services, evaluation runners, end-to-end tests, release automation, and Docker-backed MySQL and Redis setup. packages/openwork-ui-mcp/index.mjs implements a stdio MCP server that discovers a tokenized local HTTP bridge, exposes openwork://context/current, and provides context, snapshot, query, command, action, and status tools; the supplied excerpt is truncated. packages/paths/index.mjs centralizes Windows, macOS, and XDG-aware configuration and data paths, validates configuration roots, and supports environment overrides. packages/connect-link/src/index.ts and packages/install-config/src/index.ts use Zod schemas, reject insecure non-loopback transport URLs, and build connection or installation links. packages/automations/src/index.ts, packages/headless-threads/src/index.ts, and packages/enterprise-mcp-client/src/index.ts expose automation, headless conversation, and enterprise MCP client surfaces.
Observed
- License
- README describes OpenWork as free and open source; no specific license identifier appears in the supplied text.
- Primary language
- TypeScript and JavaScript modules, with TypeScript configured in the root workspace.
- Packaging
- Private pnpm 11 workspace with filtered packages, Electron development commands, Docker-backed local services, and downloadable desktop installers.
- Interfaces
- Desktop application, remote MCP endpoint, local stdio MCP server, headless thread client, and exported TypeScript packages.
- Platform support
- README states macOS, Windows, and Linux support; path handling includes Windows, macOS, and XDG conventions.
- Verification surface
- Root scripts include package tests, evaluations, specifications, health checks, permission tests, session tests, and end-to-end tests.
Read from README.md, package.json, packages/paths/index.mjs, packages/openwork-ui-mcp/index.mjs, packages/email/src/index.ts, packages/types/src/index.ts, packages/codemode/src/index.ts, packages/automations/src/index.ts, packages/connect-link/src/index.ts, packages/install-config/src/index.ts, packages/headless-threads/src/index.ts, packages/enterprise-mcp-client/src/index.ts, packages/enterprise-mcp-mock-server/src/cli.ts, packages/enterprise-mcp-mock-server/src/index.ts, packages/ui/src/react/index.ts.
What it can do
Generate text using large language models
Natural language prompts and user's API keys → AI-generated text responses
Automate browser tasks
Browser automation workflows and target websites → Executed browser actions and scraped data
Share AI workflows with team members
Created AI workflows and team collaboration settings → Shared workflows accessible to team
Connect to multiple LLM providers
API keys from various LLM services → Access to 50+ large language models
Create custom AI workflows
Workflow configuration and AI model parameters → Automated AI processing pipelines
Manage API key integrations
User's own API keys from LLM providers → Configured connections to AI services
Tags
Tech Stack
Media
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.