What is Eigent?
Eigent is a desktop app from the team behind the CAMEL-AI multi-agent framework. You describe a job; a coordinating agent breaks it into subtasks and hands them to specialist workers (developer, browser, document and multimodal agents) that run in parallel and report back. It can also run a single focused agent, schedule recurring Automations and plug in MCP servers and skills.
The front end is Electron and React; the back end is a FastAPI server running CAMEL. It works with cloud APIs, your own keys or local models through vLLM, Ollama or LM Studio, and a documented local deployment keeps accounts, settings and chat history in a Postgres database on your own machine. The repository was created in July 2025, is Apache 2.0 licensed and released v1.0.7 on 30 October 2026.
How much does Eigent cost?
| Plan | Price | What you get |
|---|---|---|
| Free | Free | Own API keys or local models |
| Plus | $19.99/month | 2,000 task credits a month |
| Pro | $99.99/month | 10,000 task credits a month |
| Enterprise | Custom |
Free and open source (Apache 2.0) with your own API keys or local models. Hosted credits: Plus $19.99/month (2,000 task credits), Pro $99.99/month (10,000 task credits), both with a 7-day trial; Teams announced; Enterprise custom.
Is Eigent worth it?
Eigent's distinguishing idea is the workforce: instead of one agent working through a long plan step by step, a coordinator assigns subtasks to specialist workers that run at the same time. For jobs with independent parts, such as researching five competitors or building several small web pages, that is a real time saving, and the project's use-case library shows it producing Word reports, chart-heavy summaries and browser games from a single prompt. It has about 15,800 GitHub stars and has shipped a release roughly every two to three weeks since September 2026, most recently v1.0.7 on 30 October.
This is a computer-use agent in the full sense. The developer worker runs terminal commands, the browser worker drives a real browser, and the agent prompts tell it to control native applications with AppleScript on macOS and pyautogui (synthetic mouse clicks and keystrokes) elsewhere. Four approval modes govern this, from Read only and Ask first to an automatic reviewer and Full access. The project's own SECURITY.md is blunt that these approvals and the audit trail are not an operating-system sandbox, and that an agent with a shell has the run of your files and processes.
The security history argues for caution. Two advisories are published: a high-severity 1-click remote code execution flaw in the app (December 2025) and a critical CI workflow weakness that let fork pull requests run code with write access to the repository (January 2026). Both are now public, and both are early and serious for a project of this age.
Privacy depends on how you run it. The quick start connects to Eigent's cloud and needs an account; the local deployment keeps registration, settings and history in your own Postgres instance. The privacy policy says Eigent does not collect prompts or files from local deployments, mentions optional error reporting an administrator can switch on, and reserves the right to use anonymised derived data from its cloud service. Compared with OpenWork, Eigent is stronger at parallel multi-agent work and weaker on team administration; compared with Goose it is friendlier for non-developers but far less proven.
Where it earns its place
An engineering manager at a 30-person software company
Turn a month of merged pull requests into a written report and a Slack update
Eigent's own guide runs this with DeepSeek through Ollama: the agents read the GitHub pull requests, write a Word summary and draft the release note. With a local model the software cost is nil; the cost is a machine able to run the model.
A market analyst at a small consultancy
Research several competitors at once and assemble one comparison document
The coordinator assigns one browser worker per competitor and a document worker to merge the findings. On hosted credits, Plus at $19.99 a month includes 2,000 task credits; heavy multi-agent days are better served by your own API key.
A data protection officer at a regulated firm
Trial a desktop agent without sending staff files to a vendor cloud
Run the local deployment with its own Postgres database and a local model, set approvals to Ask first, and keep optional error reporting off. Treat the agent's shell and GUI access as full user access when deciding which machine it runs on.
Before you commit
- Setup
- Download the desktop app from eigent.ai, or from source clone the repo with Node 18 to 22 and run npm install and npm run dev; that quick start uses Eigent's cloud and needs an account. The fully local route follows the server guide and adds a local Postgres database.
- Integrations
- Cloud model APIs, enterprise gateways and local models through vLLM, Ollama or LM Studio; MCP servers and skills; built-in browser, terminal and document toolkits; scheduled Automations; remote sub-agents such as BytePlus ModelArk.
- Data and privacy
- Local deployment keeps registration, settings and chat history on your machine; the privacy policy says prompts and files from local deployments are not collected, with optional error reporting an admin can enable. Cloud accounts store account and usage data, and Eigent may use anonymised derived data. Agents run shell commands, drive a browser and can move the mouse and type through pyautogui or AppleScript with your user's permissions.
- Cost at scale
- Self-hosting is free plus model costs. Hosted credits are Plus $19.99 a month for 2,000 task credits or Pro $99.99 a month for 10,000; a Teams plan is announced but not priced, and Enterprise with SSO and local deployment is quoted.
- Limits
- Approvals and audit trail are not an OS sandbox, by the project's own account. Two published advisories, one critical. About 248 open issues and pull requests. Multi-agent runs multiply token use.
What’s good
- Parallel specialist agents for tasks with independent parts
- Local deployment keeps accounts and history on your machine
- Works with local models through vLLM, Ollama or LM Studio
- Four approval modes, from Read only to Full access
What’s not
- Shell and GUI control run with your full user rights, no sandbox
- A high-severity 1-click RCE and a critical CI flaw already published
- Quick start ties you to a cloud account
- Multi-agent runs burn through credits or tokens quickly
Best for
Technical individuals and small teams who want parallel agents for research, reporting and file work on their own machine, and are willing to supervise them.
Eigent alternatives
Eigent: common questions
- Is Eigent worth it?
- Eigent is the most ambitious of the open Cowork clones, splitting work across parallel agents that can drive your desktop, but its shell and GUI access run with your full user rights and its security record is short and already includes a 1-click RCE. We rate it 3.9 out of 5.
- How much does Eigent cost?
- Free and open source (Apache 2.0) with your own API keys or local models. Hosted credits: Plus $19.99/month (2,000 task credits), Pro $99.99/month (10,000 task credits), both with a 7-day trial; Teams announced; Enterprise custom.
- What are the downsides of Eigent?
- Shell and GUI control run with your full user rights, no sandbox; a high-severity 1-click RCE and a critical CI flaw already published; quick start ties you to a cloud account; multi-agent runs burn through credits or tokens quickly.
- Who is Eigent best for?
- Technical individuals and small teams who want parallel agents for research, reporting and file work on their own machine, and are willing to supervise them.
- How much does Eigent cost for a team or in production?
- Self-hosting is free plus model costs. Hosted credits are Plus $19.99 a month for 2,000 task credits or Pro $99.99 a month for 10,000; a Teams plan is announced but not priced, and Enterprise with SSO and local deployment is quoted.
- Is Eigent safe for company data?
- Local deployment keeps registration, settings and chat history on your machine; the privacy policy says prompts and files from local deployments are not collected, with optional error reporting an admin can enable. Cloud accounts store account and usage data, and Eigent may use anonymised derived data. Agents run shell commands, drive a browser and can move the mouse and type through pyautogui or AppleScript with your user's permissions.
- What are the best Eigent alternatives?
- The closest alternatives we have reviewed are OpenWork, Interpreter Workstation, OpenWorker, goose and Magentic-UI (MagenticLite).
Answers checked Nov 14, 2026.