cua
Computer-use platform (Cua Spaces, Driver, Lume) with CUA-S1 small decision models and a preview jev-use example for bounded actions.
Overview
Cua is a platform for giving AI agents real computers: Cua Spaces provides macOS and Linux desktops for agents, Cua Driver handles desktop automation, Lume runs local VMs, and Cua Bench evaluates computer-use agents. The repository also holds CUA-S1, a family of small specialised System 1 models for fast, bounded computer-use decisions such as which value belongs in a field or whether to leave an element alone.
The decision-model parts are CUA-S1, an early source-only research release with weights on Hugging Face, and, according to the list summary, a preview jev-use example that pairs Driver observation and execution with bounded Jev browser-action choices. Application code orders and executes the actions; the decision models only score options. Most of the repository is general computer-use infrastructure rather than decision-model code.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which value from the document belongs in this form field? | — |
| Choice | Which bounded browser action should be taken next?(inferred) | — |
How it works
- Cua Driver observes the interface as structured elements
- A CUA-S1 model or Jev scores bounded options
- Application code orders the actions
- Driver executes with explicit action boundaries
Strengths
- Small specialised models for frequent, bounded decisions
- Full computer-use stack (VMs, driver, benchmark) around the decision step
Caveats
- CUA-S1 is an early research release
- Jev integration is a preview example
- Decision models are a small part of a large repository
Getting started
curl -fsSL https://cua.ai/install.sh | sh on macOS 26 or later; see libs/cua-s1 for the decision models.At a glance
- Type
- Project
- Status
- beta
- Language
- Python
- Runtime
- macOS app, CLI, SDK
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Cua’s preview jev-use example pairs Driver observation and execution with bounded Jev browser-action choices.
logicrw/awesome-jev-projects
Frequently asked questions
What is CUA-S1?
A family of small System 1 models for computer-use decisions, starting with form filling.
Where does Jev come in?
A preview jev-use example uses Jev for bounded browser-action choices with Cua Driver.
Is the whole repo about decision models?
No, most of it is computer-use infrastructure such as Spaces, Driver and Lume.
Related projects
706K-parameter Jev-style option scorer that rates fill, check, click or skip for each form element in one parallel pass.
ModelresearchPythonLocal browser agent that picks each step's operation and element with a Jev-shaped model such as a fine-tuned Laya, no cloud needed.
MCP serverbetaPythonLets coding agents run whole goals in your signed-in Chrome on macOS in background tabs, asking Jev or Claude Haiku only for real choices.
ProjectbetaLocal stdio MCP server that executes whole browser and macOS goals with a Jev-driven observe-decide-act loop.
MCP serverbetaTypeScriptBrowser Use's browser agent where Jev picks an operation and page element in one request and a small LLM only writes text to type.
LibrarybetaPythonPython decision layer for browser and desktop automation: code observes, acts and verifies; the model makes one typed choice per step.
ProjectbetaPython