otto
Desktop computer-use agent for macOS and Windows where Jev chooses each concrete native action, with optional GPT-6 Astra planning.
Overview
Otto is an open-source desktop computer-use agent for macOS and Windows that works through native accessibility controls (macOS Accessibility, Windows UI Automation) plus local OCR for text missed by accessibility. In Jev-only guided tasks you select apps and describe a task, and Jev chooses concrete actions such as press, set text, scroll or bounded keys; each exact action is shown for approval before it executes, and the next observation checks what changed. Hybrid mode adds GPT-6 Astra for plans and drafted text between Jev's action choices.
A fill-a-form workflow applies a literal field map after one review with no model calls, verifying each write by native readback. An agent bridge lets Codex or another MCP host use scoped native inspection and actions, with optional bounded Jev delegation. The project is an early alpha: broad app compatibility, model-driven task success and signing are unverified, and model confidence is shown as a score, not a guarantee.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which concrete native action should be taken next toward the goal? | — |
How it works
- Read selected apps via accessibility and local OCR.
- Jev picks the next concrete action (planner adds subgoals in Hybrid).
- Show the exact action for approval, then execute.
- Re-observe to confirm the change; user confirms the result.
Strengths
- Every action is reviewed before execution.
- Form fill mode needs no model calls.
- Native controls and local OCR.
Caveats
- Early alpha; app compatibility unverified.
- Hybrid mode needs an OpenAI Platform key as well.
Getting started
Download an alpha build from GitHub Releases and add a TypeSafe key for Jev-only guided tasks.At a glance
- Type
- Project
- Status
- prototype
- Runtime
- desktop app (macOS, Windows), MCP bridge
- Has code
- yes
Use cases and integrations
How curated lists describe it
Open-source native computer use for macOS and Windows: TypeSafe Jev, local OCR, and selective planning.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev decide in Otto?
The next concrete native action, such as pressing a control or setting a field.
Does it act without asking?
No, guided tasks show each exact action for approval first.
Do I need OpenAI?
Only for Hybrid mode; Jev-only mode needs just a TypeSafe key.
Related projects
Accessibility-first macOS computer-use MCP server and CLI, with optional Jev guards on targets and inputs before irreversible actions.
PluginbetaPythonCodex computer-use skill where Jev picks the next on-screen element, action, completion and risk from text candidates.
Agent skillprototypeJavaScriptVerified computer-use agent: a Jev guard checks claimed completions and gates irreversible actions; Jev also takes simple steps.
ProjectbetaPythonMCP server that lets agents observe and control Android phones over ADB, with Jev picking controls and judging goal completion.
MCP serverbetaTypeScriptVoice assistant daemon for Hyprland Linux desktops that acts through tool modules, with Jev gating which requests and actions proceed.
ProjectbetaPythonRust harness for desktop and browser automation where Jev answers small closed questions about the screen to pick each action.
ProjectbetaRust