Open Computer Use (E2B cloud desktop)

An open-model agent that operates an isolated cloud desktop through keyboard, mouse, and shell actions.

Stars
★ 2.3k
Last updated
2mo ago
License
Apache-2.0
Primary language
Python

At a glance

Works with
Platform-specificOpenAI API · Claude API
You'll need
Python 3.10 or latergitPoetryffmpegE2B API keyLLM provider API keyShell / CLINetwork access
Typical use
A developer who needs an LLM to operate keyboard, mouse, and shell controls inside an isolated cloud Ubuntu desktop.
Main limitation
Core operation depends on E2B Desktop Sandbox and requires an E2B API key.

What does this agent do, and when should you use it?

Open Computer Use is a cloud Linux computer powered by E2B Desktop Sandbox and controlled by open-source LLMs. The agent operates the sandbox through keyboard, mouse, and shell commands, while its display is live-streamed to the client computer. It includes a web interface where users can pause the agent and provide another prompt at any time. Ubuntu is the documented operating system, although the project says the design is intended to work with other operating systems. Model selection is configured in os_computer_use/config.py, with provider integrations defined in providers.py.

After receiving a prompt, the agent controls the cloud Linux computer supplied by E2B Desktop Sandbox using keyboard, mouse, and shell commands, and streams the sandbox display to the client. A user can pause execution and send an additional prompt. os_computer_use/config.py selects the models, for example with providers.OSAtlasProvider() for grounding and providers.GroqProvider("llama3.2") or providers.GroqProvider("llama3.3") for vision and actions. The documented providers in providers.py include Fireworks, OpenRouter, Llama API, Groq, DeepSeek, Google, OpenAI, Anthropic, HuggingFace Spaces, Moonshot, and Mistral AI.

  1. A developer who needs an LLM to operate keyboard, mouse, and shell controls inside an isolated cloud Ubuntu desktop.
  2. An operator who wants to watch the remote desktop live and interrupt the agent when a task needs correction.
  3. A team evaluating Groq, OpenAI, Anthropic, Gemini, or another documented provider by changing model configuration in config.py.
  4. A researcher experimenting with OS-Atlas or ShowUI as the agent's grounding model.
  5. A user who wants to start the web interface with an initial CLI prompt, such as asking the agent to use a web browser for a weather lookup.

How do you install or deploy this agent?

Prerequisites are Python 3.10 or later, git, an E2B API key, and an API key for the selected LLM provider. Install Poetry and ffmpeg with: brew install poetry ffmpeg. Clone the project with: git clone https://github.com/e2b-dev/open-computer-use/. Enter it with: cd open-computer-use. Create a .env file containing E2B_API_KEY="your-e2b-api-key", then add the key required by the provider selected in config.py, such as GROQ_API_KEY, OPENAI_API_KEY, or ANTHROPIC_API_KEY. For HuggingFace Spaces, the documented setup requires HF_TOKEN to bypass Gradio rate limits. Run: poetry install.

How do you use this agent?

From the project directory, run: poetry run start. This starts the web interface, which opens and asks for the first instruction. To supply one at launch, run: poetry run start --prompt "use the web browser to get the current weather in sf". The sandbox display stream should appear a few seconds after the Python program starts. Change grounding_model, vision_model, and action_model in os_computer_use/config.py as needed, and ensure the matching provider credentials are in .env.

What are this agent's strengths and limitations?

Pros
  • Uses E2B Desktop Sandbox for an isolated cloud Linux desktop and live-streams that desktop to the client.
  • Separates grounding, vision, and action model configuration, enabling documented provider and model combinations in config.py.
  • Documents grounding support for OS-Atlas and ShowUI alongside multiple API-based model providers.
  • Users can pause the agent and send another prompt while it is operating the computer.
Limitations
  • Core operation depends on E2B Desktop Sandbox and requires an E2B API key.
  • Except for HuggingFace Spaces, selected model providers require separate API credentials, adding account, billing, and secret-management overhead.
  • The documented setup uses brew and requires Poetry and ffmpeg; equivalent instructions for other installation environments are not supplied.
  • Ubuntu is the explicitly documented operating system; support details for other operating systems are not provided.

How does this agent compare with similar options?

Key facts side by side with the most closely related agents.

Agent Source review Stars Updated Language Full support on
Open Computer Use (E2B cloud desktop) This agent 0 · Major gaps ★ 2.3k 2mo ago Python OpenAI API · Claude API
Clawd Cursor 74 · Some gaps ★ 403 today TypeScript Codex · Claude Code
Hermes Agent 67 · Some gaps ★ 248k today Python OpenAI API
macOS Harness 55 · Major gaps ★ 886 1mo ago Python Codex · Claude Code

How does FollowAgents rate this agent?

FollowAgents source review · FARS-2.1
Major gaps
0/ 100 5-point scale 0.0 / 5
Trust 0/29
Reliability 0/14
Adaptability 0/18
Convention 0/18
Effectiveness 0/13
Verifiability 0/8
Why each dimension lost points
Trust0 / 29 · 0.0/5

Insufficient evidence: The repository provides no documentation or implementation for least privilege, user confirmation, data flow transparency, sensitive data handling, dependency security, external effects, rollback, or source attribution. All criteria deducted for lack of evidence.

Reliability0 / 14 · 0.0/5

Insufficient evidence: No documentation or implementation for self-consistency, dependency availability, or failure messages. All criteria deducted for lack of evidence.

Adaptability0 / 18 · 0.0/5

Insufficient evidence: No documentation for audience, capability boundaries, trigger precision, or environment fit. All criteria deducted for lack of evidence.

Convention0 / 18 · 0.0/5

Insufficient evidence: Information architecture, install notes, naming stability, examples/FAQ, known limitations, license, versioning/changelog, and maintenance responsibility are not adequately documented. License file exists but not clearly stated in README. All criteria deducted for lack of evidence.

Effectiveness0 / 13 · 0.0/5

Insufficient evidence: No documentation for output usability, marginal value, or cost-benefit. All criteria deducted for lack of evidence.

Verifiability0 / 8 · 0.0/5

Insufficient evidence: No documentation for claim traceability, cross-source corroboration, or fact/inference separation. All criteria deducted for lack of evidence.

Risks and how to mitigate them
  • Not found in source: least-privilege scopingGrant only what the task needs: a dedicated account or read-only token, scoped to specific directories and repos.
  • Not found in source: confirmation before actingTurn on (or add) a confirmation step before it acts, and try it in a sandbox or test environment before real data.
  • Not found in source: data-flow disclosureWatch which external services it contacts (proxy or firewall logs) and keep sensitive data out until you know where it goes.
  • Not found in source: sensitive-data handlingUse dedicated, low-privilege, revocable API keys — never production credentials — and keep secrets out of logs.
  • Not found in source: dependency securityPin versions and run a dependency audit (npm audit, pip-audit) before installing; prefer running it in a container.
  • Not found in source: disclosed external effectsEstablish which external systems it writes to, sends to or changes, and verify with test accounts or repos before production.
  • Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
  • Not found in source: verifiable attributionInstall from the official repo or registry and check the publisher and URL to avoid look-alike packages.
  • The repository lacks security or permission control documentation; assess risks before use.
  • No user confirmation mechanism is provided; the agent may execute actions without confirmation.
  • Dependencies are not security-audited; potential vulnerabilities exist.
  • No rollback or recovery mechanism is provided; failed actions may not be reversible.
Evidence confidence: Low Reviewed Aug 09, 2026 Reviewed revision 610bac85d242
See the full review method →

FAQ

Which credentials are required?
An E2B_API_KEY is required, along with the API key for the provider selected in config.py. HuggingFace Spaces does not require a provider API key, but the documented setup requires HF_TOKEN to bypass Gradio rate limits.
What can the agent control?
The documented controls are keyboard, mouse, and shell commands, applied to a cloud Linux computer running in E2B Desktop Sandbox.
Can I intervene while it runs?
Yes. The user can pause the agent at any time and provide another prompt.
Can I use OpenAI or Anthropic models?
Yes. The documented providers include OpenAI GPT-4o and GPT-4o mini, plus Anthropic Claude. Select the provider in configuration and supply its API key.
View on GitHub ↗ Install ↓

Related agents