tinycomputer
Rust harness for desktop and browser automation where Jev answers small closed questions about the screen to pick each action.
Overview
tinycomputer automates desktop apps such as Mail and Notes, and websites in a real Chrome browser. It reads the screen through accessibility data, as a list of named buttons and fields, rather than screenshots. Every choice about what to press comes from Jev answering yes/no, scale or which-option questions; the Rust harness reads the screen, asks the questions, acts, verifies the result and enforces safety rules.
Callers hand it a plain-language task, and it runs in the background, pausing for missing details, approval before irreversible actions, captchas or one-time codes, and always stopping at payment. Each question is asked several ways with shuffled options and varied wording, then averaged to cancel position and yes-bias; unclear results trigger more questions and head-to-head comparisons. A reasoning model suggests alternatives when a step is stuck. It ships as a TinyBus module.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which on-screen element should be pressed for this step?(inferred) | — |
| Yes / no | Did the last action move the task forward?(inferred) | yes · no |
| Yes / no | Is this action irreversible (send, delete, submit)?(inferred) | yes · no |
How it works
- Read the accessibility tree of the browser or app.
- Ask Jev several varied, option-shuffled questions and average the answers.
- Act on the winning element, then check what changed.
- Undo and try the runner-up if things got worse; pause for approval, input or payment.
Strengths
- No screenshots or pixel matching; uses accessibility data.
- Averaging shuffled variants reduces position and yes-bias.
- Stops before payment and asks before irreversible actions.
- Secret details are typed into fields without being shown to a model.
Caveats
- Depends on a planner and a reasoning model in addition to Jev.
- Platform support and test coverage are not detailed in the excerpt.
At a glance
- Type
- Project
- Status
- beta
- Language
- Rust
- Runtime
- TinyBus module (library loaded by a host)
- Has code
- yes
Use cases and integrations
How curated lists describe it
Decision model (JEV) based harness to perform Desktop & Browser automation.
wh000wh000/awesome-jev-live
Frequently asked questions
What does Jev decide in tinycomputer?
Which button or field to act on and whether actions worked, through small yes/no, scale and option questions.
Does it use screenshots?
No, it reads the screen as a screen reader would, as named controls.
Will it spend money?
No, it stops at the payment step and leaves the page open for a person.
Related projects
Claude Desktop MCP extension that drives Brave or Chrome with Jev choosing on-page actions and a guard freezing irreversible clicks.
MCP serverbetaTypeScriptChrome MCP server with 50 tools that drives your everyday browser without losing logins, where Jev or a local decider picks each step.
MCP serverbetaLets coding agents run whole goals in your signed-in Chrome on macOS in background tabs, asking Jev or Claude Haiku only for real choices.
ProjectbetaComputer-use platform (Cua Spaces, Driver, Lume) with CUA-S1 small decision models and a preview jev-use example for bounded actions.
ProjectbetaPythonLocal stdio MCP server that executes whole browser and macOS goals with a Jev-driven observe-decide-act loop.
MCP serverbetaTypeScriptVerified computer-use agent: a Jev guard checks claimed completions and gates irreversible actions; Jev also takes simple steps.
ProjectbetaPython