Interpreter Workstation

An open, neutral AI workspace for your computer.

Interpreter Workstation is a desktop agent that works with you across files, documents, spreadsheets, PDFs, the browser, and your computer. Built on Open Interpreter, it can use your choice of supported cloud provider or local model, and it does not require an Interpreter account.

It is a self-hostable alternative to proprietary work agents such as Claude Cowork and ChatGPT, without making another vendor's product the definition of what this project can become.

Watch the 95-second launch film →

Bring your own model, keep work local when you choose, and control which files and applications each agent may access. Organizations can build a configured distribution from the same source without maintaining a private fork.

Architecture

• Open Interpreter owns the agent runtime, harnesses, provider/model discovery, and app-server protocol.
• This repository owns the Electron shell, user experience, local policy, and workstation tools.
• Model-facing app tools use the interpreter-app command-line contract. The desktop app does not maintain a second provider or harness implementation.
• The browser extension and computer-use driver are pinned Git submodules so a desktop release is reproducible while their independent release histories are preserved.

Requirements

• Node.js 22 (see .nvmrc)
• pnpm 9
• Bun
• Rust stable and Cargo
• Git with submodule support

Build from source

git clone --recurse-submodules https://github.com/openinterpreter/interpreter-workstation.git
cd interpreter-workstation
pnpm install
pnpm run download:oix -- --current-platform
pnpm run download:pdfcpu -- --current-platform
pnpm run download:qwen-asr -- --current-platform
pnpm run build
pnpm start

The default product configuration is the account-optional community distribution. Its hosted account, hosted API, telemetry, and external document engine fields are empty.

Development

pnpm dev
pnpm typecheck
pnpm run test:unit
pnpm run test:vitest
pnpm test

pnpm test also builds the app and runs Electron end-to-end coverage. Voice and live-provider tests are opt-in because they require platform assets or external services.

The browser extension can be bootstrapped and verified independently:

pnpm run extension:bootstrap
pnpm run extension:verify

OIX runtime and shared home

Workstation installs and launches the exact checksummed OIX release pinned by the app. Terminal OIX uses an independently movable current selector, so a terminal update cannot change the app-server protocol underneath a released Workstation build. If no terminal installation exists, Workstation exposes its pinned release as both interpreter and i; later terminal-managed updates can take ownership of that selector without changing the app runtime.

The app and terminal share one OIX home: INTERPRETER_HOME when set, otherwise ~/.openinterpreter. Configuration, sessions, plugins, and global skills therefore work across both surfaces. Workstation installs and updates only the skills it ships, preserves user and enterprise skills, and backs up local edits before replacing a managed skill. OIX separately owns and updates its embedded .system skills. See Skills and the Open Interpreter home and local OIX testing.

Distributions

Product-specific hosted services are configuration, not a separate application. The official hosted profile and its release configuration are public and live in distribution/; no private client fork is required. Use the distribution wrapper for an organization-specific JSON overlay:

node scripts/with-distribution-config.mjs ./path/to/product.overlay.json -- pnpm run build

See Distribution builds for the schema, security boundary, privacy contract, and community/official/internal packaging model. See Official releases for the signing, approval, provenance, checksum, SBOM, and verification contract.

Document workflows

Interpreter's primary document workflow is code execution plus skills. A compatible external document engine can optionally provide rich embedded editing and format conversion. It is not bundled with the community source. See Document engines.

Enterprise deployment

The application is designed to be deployable as an organization's primary AI workstation: account-optional, provider-configurable, policy-aware, and buildable from one canonical source tree. See Enterprise deployment for the current controls and the remaining production-hardening checklist.

Security

Please report vulnerabilities privately as described in SECURITY.md. Do not put credentials in product overlays or commit local .env files.

License

Interpreter Workstation is licensed under the Apache License 2.0. Project-authored documentation is available under CC BY 4.0. See GOVERNANCE.md for the one-open-client and no-relicense commitments. Official-release provenance and use of project marks are described in TRADEMARKS.md. Pinned dependencies and submodules retain their own licenses and notices. Read Dependency licensing and the reviewed third-party notices before distributing a packaged binary. Every packaged app carries these notices under its licenses/ resource directory.