Computer Use logo

Computer Use

Organization
vm0-ai
computer-use

Operate apps on a desktop host through Computer Use, whether surfaced as Okou Desktop or the still-supported legacy Zero Desktop, when APIs are not enough. Not for remote browser sessions (`okou browser use`), surfaced as Okou Browser or legacy Zero Browser.

Overview

Publishervm0-ai
Repositoryvm0-skills
Skill namecomputer-use
Stars
76
Forks
18
Bundled files
Instructions only
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • Self-contained

    Everything the model needs lives in the instructions — no extra files to sync.

  • Open source

    Published by vm0-ai on GitHub. Read the source before you install it.

Installation

Install the Computer Use AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/vm0-ai/vm0-skills.git /tmp/vm0-skills
mkdir -p .claude/skills
cp -r /tmp/vm0-skills/computer-use .claude/skills/computer-use
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Computer Use in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Computer Use on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Computer Use is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Computer Use

Overview

Use okou computer-use <command> to inspect and operate apps on the connected desktop host. Current clients present this as Okou Desktop; still-supported legacy clients may call it Zero Desktop. Treat it as an accessibility-first GUI control surface: read the app state, act through accessibility elements, and inspect screenshots only when visual information is required or the accessibility state is insufficient.

--app accepts an app bundle id only, such as com.apple.Safari or com.google.Chrome. App names like Safari, Google Chrome, Slack, or WeChat are display labels only and must not be passed to --app.

Core Workflow

  1. List apps and choose the target app's bundleId:
bash
okou computer-use list-apps

Use the name field only to identify the app for yourself, then copy the bundleId field. Apps listed without a bundleId cannot be targeted.

  1. Set the bundle id you found:
bash
app_bundle_id="com.apple.Safari"
  1. If the app is not running, open it by bundle id:
bash
okou computer-use open-app --app "$app_bundle_id" --timeout 10
  1. Inspect the target app by bundle id:
bash
okou computer-use get-app-state --app "$app_bundle_id" --timeout 10
  1. Read the JSON result:
  • snapshotId is required for follow-up actions against the same state.
  • appState is expected to be a local file path containing the full accessibility tree. Read or filter that file to find element indexes, roles, labels, URLs, and the focused element.
  • screenshot is a local image path under /tmp/vm0/computer-use/.... Do not inspect it by default; use the image viewer only when visual judgment matters or the accessibility tree does not expose a usable target.
  • During rollout, older CLI versions may still return appState as an inline string. If it is not a path, save the JSON output to a temp file and extract/filter the inline string.
  1. Act on elements first, using the same bundle id and snapshot:
bash
okou computer-use click --app "$app_bundle_id" --snapshot-id <snapshotId> --element-index <index>
okou computer-use set-value --app "$app_bundle_id" --snapshot-id <snapshotId> --element-index <index> --value "text"
okou computer-use type-text --app "$app_bundle_id" --snapshot-id <snapshotId> --text "literal text"
okou computer-use press-key --app "$app_bundle_id" --snapshot-id <snapshotId> --key Command+L
okou computer-use scroll --app "$app_bundle_id" --snapshot-id <snapshotId> --element-index <index> --direction down --pages 1
  1. Re-read state after every meaningful UI change. Element indexes and coordinates can become stale after navigation, scrolling, or opening a new window.

If a post-action command returns a new snapshotId and appState, use that new snapshot. Otherwise, explicitly run get-app-state again before choosing the next element index or coordinate.

Accessibility-First Policy

  • Start with the appState accessibility tree. For routine forms, buttons, lists, links, and navigation, do not inspect the screenshot when accessibility roles and labels provide enough information to act safely.
  • If the UI behaves unexpectedly, re-read and filter the latest appState before falling back to the screenshot.
  • Inspect the screenshot when the accessibility tree is incomplete or ambiguous, the task depends on visual appearance or layout, or no useful accessibility element exists and coordinate input is required.
  • Use coordinates only after inspecting the screenshot from the same snapshot. Return to element-index actions as soon as the accessibility tree exposes a reliable target.

App Identity Rules

  • Always run list-apps before targeting an app unless you already know the exact bundle id from the same host.
  • Always pass the bundleId value to --app.
  • Never pass display names such as Slack, Safari, Google Chrome, or WeChat to --app.
  • If a command fails with app_not_found and the --app value looks like a name, re-run list-apps, find the app record, and retry with its bundleId.
  • If the target app has no bundleId in list-apps, report that it is not targetable through Computer Use.

To search the list-apps output for a likely target:

bash
apps_json=/tmp/computer-use-apps.json
okou computer-use list-apps > "$apps_json"
node -e "const fs=require('fs'); const q=(process.argv[2]||'').toLowerCase(); const apps=JSON.parse(fs.readFileSync(process.argv[1],'utf8')).apps || []; for (const a of apps) { if (!q || String(a.name || '').toLowerCase().includes(q)) console.log([a.name || '', a.bundleId || 'NO_BUNDLE_ID', a.running ? 'running' : 'not running'].join('\t')); }" "$apps_json" "Slack"

Copy the exact bundleId from the matching row into app_bundle_id.

Reading App State

Accessibility trees can be very large. Prefer the appState file returned by the command:

bash
app_bundle_id="com.apple.Safari"
state_json=/tmp/computer-use-state.json
okou computer-use get-app-state --app "$app_bundle_id" --timeout 10 > "$state_json"
app_state_path=$(node -e "const fs=require('fs'); const r=JSON.parse(fs.readFileSync(process.argv[1],'utf8')); console.log(r.appState)" "$state_json")
rg -n "Ethan|showcase|sites\\.vm0|Send now" "$app_state_path"

If the CLI still returns inline app state instead of a path, use the transition workaround:

bash
state_json=/tmp/computer-use-state.json
okou computer-use get-app-state --app "$app_bundle_id" --timeout 10 > "$state_json"
node -e "const fs=require('fs'); const s=JSON.parse(fs.readFileSync(process.argv[1],'utf8')).appState; s.split('\\n').forEach((l,i)=>{ if(/Ethan|showcase|sites\\.vm0|Send now/.test(l)) console.log((i+1)+': '+l) })" "$state_json"

Use filtered output to locate clickable link, button, text field, checkbox, or outline row entries and their element indexes.

Action Rules

  • Use the same bundle id for get-app-state and follow-up actions against that snapshot.
  • Prefer --element-index or --element over coordinate clicks.
  • Use --x/--y only when the screenshot clearly shows the target but the accessibility tree has no useful element.
  • Always pass the matching --snapshot-id from the state used to pick the element or coordinate.
  • Use open-app --app <bundleId> when the app is not visible or not running.
  • Use press-key for reliable shortcuts such as Command+L, Command+T, Command+W, Enter, and arrow keys.
  • Use type-text only after confirming focus is in the right field.
  • If type-text fails with element_not_editable, click into a text field first or use set-value on a settable text field.
  • Avoid destructive UI actions unless the user explicitly requested them and the current state confirms the target.

Browser Pattern

After clicking a link from another app:

  1. Use list-apps to find the browser bundle id, such as com.apple.Safari, com.google.Chrome, or the Arc bundle id returned by the host.
  2. Read or filter the browser appState for the address field, page title, links, controls, and focused element.
  3. Use press-key --app <browserBundleId> --snapshot-id <snapshotId> --key Command+L plus type-text and Enter when direct navigation is faster than finding a link.
  4. Inspect the screenshot only when the task depends on layout, design, images, canvas content, or another detail that the accessibility tree does not represent reliably.

Failure Handling

  • app_not_found: check whether --app was a display name. If so, retry with the app's bundleId from list-apps. If it was already a bundle id, the app may not be installed, not running, or not targetable on the host.
  • window_unavailable: try open-app --app <bundleId>, then run get-app-state again. If it still fails, report that the app has no controllable window.
  • element_not_editable: click into the target text field, re-read state, then use type-text; or use set-value when a settable text field is available.
  • permission_denied, accessibility_unavailable, or screen_recording_unavailable: report the missing Desktop permission instead of retrying the same command.

Reporting Back

State what was actually done through Computer Use: apps inspected by bundle id, whether screenshot fallback was needed, and whether actions succeeded. If a desktop action times out or the host is unavailable, say that directly and use a stable fallback only when it still satisfies the user request.

Frequently asked questions

What does the Computer Use AI skill do?

Operate apps on a desktop host through Computer Use, whether surfaced as Okou Desktop or the still-supported legacy Zero Desktop, when APIs are not enough. Not for remote browser sessions (`okou browser use`), surfaced as Okou Browser or legacy Zero Browser.

Why use Computer Use on TypingMind?

Because you install it once and use it with any model. Computer Use is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Computer Use in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/vm0-ai/vm0-skills/tree/main/computer-use. TypingMind reads its SKILL.md and installs it as a skill you can enable per chat.

Which AI models can use Computer Use?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Computer Use?

As many as you like. As long as a model supports skills, you can use Computer Use with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Computer Use AI skill free?

It is published on GitHub by vm0-ai. Check the repository for licensing terms. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇