MISSION CONTROL — GROK × xAI · 100% LOCAL

Your Grok agent studio.
On your machine.

Build, orchestrate, and schedule AI agents with full computer use — chat that acts, a sub-browser that annotates your app, and automations that run while you sleep. Powered exclusively by Grok / xAI. Nothing leaves your box.

⬇ Install Shiba Studio ★ View on GitHub
$ git clone https://github.com/stevologic/shiba-studio.git && cd shiba-studio && npm install && npm run dev
Windows · macOS · Linux — Node.js ≥ 22.5, nothing to compile
/git pr/annotate/workspace /search/fetch/remember /recall/note/x/help
you ▸ /workspace C:\dev\pricing-app # bind this chat to the repo 📁 Chat workspace set — file tools and /git now run there you ▸ /git checkout feat/pricing-page ✓ Created and switched to new branch feat/pricing-page you ▸ /annotate # highlight the hero div on localhost:5173 you ▸ Make this container a responsive 2-column grid 📎 annotation.png grok ▸ Done — refactored .hero-grid with container queries… you ▸ /git pr Responsive hero grid | Refactor per annotation ✓ Pull request #42 opened: feat/pricing-page → main

See it in action

Real screenshots from a running install — the demo reel advances on its own.

localhost:3000
Grok Chat: annotated sub-browser element request with selector, HTML snippet, and mic-enabled composer Dashboard: hero, readiness badges, quick stats, and recent agent runs with statuses and view-answer links Automations: scheduled agents with Active/Paused states, per-schedule Run/Edit/Delete, and compact run-log icon
30+built-in agent tools
6core integrations
4model routes: token · OAuth · CLI · local
100%local — zero telemetry
AES-256credentials at rest

An engineering copilot with hands

Everything a frontier chat client does — plus the ability to actually do things.

💬

Chat that acts

Streaming reasoning, markdown & code, inline images, multimodal attachments, per-chat workspace folders — and the full slash-command deck with autocomplete:

  • /git status branch, changed files, recent commits
  • /git checkout <branch> switch to a branch, or create it
  • /git commit <message> stage everything and commit
  • /git pr <title> | <body> push and open a GitHub PR
  • /annotate highlight an element for refinement
  • /workspace <path> bind the chat to a repo/folder
  • /search <query> web search into the conversation
  • /fetch <url> read any page as clean text
  • /remember <key> | <content> save a persistent memory
  • /recall [keyword] list saved memories
  • /note <path> | <content> create an Obsidian note
  • /x <text> post to X through the integration
  • /help the full reference, in chat
🎯

Annotation sub-browser

Load the app you're building, click any element DevTools-style, and send its selector + HTML + highlighted screenshot into chat for pinpoint code refinement.

🤖

Autonomous agents

Each with its own model, git worktree, integration scopes, skills, peers, and persistent memory. 30+ built-in tools: files, shell, browser, web research, image generation, PRs.

⏰

Headless automations

Cron-scheduled runs with live execution traces. Schedules arm at server start — agents keep working with the browser closed. Deleted agents retire their own schedules.

🔌

Capabilities

GitHub, Slack, Google Drive, Discord, X, Obsidian, MCP servers, and custom skills. Scoped per-agent — an Obsidian-scoped agent knows its whole vault.

🔒

Local-first, audited

Credentials AES-256-GCM encrypted at rest. Runs, chats, and every action in an embedded SQLite audit log. Zero telemetry — traffic goes only to xAI and services you connect.

Point at a div. Ask for the fix. Ship the PR.
The annotation sub-browser + /git pr turn "that container looks wrong" into a merged pull request — without leaving chat.

How you use it

Three minutes from clone to a working agent.

Connect a model source

Paste an xAI API key, sign in with X (OAuth 2.0), point at a local model server (Ollama, LM Studio), or let it detect the Grok CLI. Readiness badges show what's live.

Chat, annotate, refine

Talk to Grok with your uploads and projects as context. Open the sub-browser, highlight the exact div that bothers you, and ask for the fix.

Automate it

Wrap the workflow in an agent, give it scopes and a schedule, and watch live traces on the Automations page — or don't watch; it runs headless either way.

Much agents. Very local. Wow.
Your models, your machine, your data. Agents that keep working after you close the browser — and remember what they learned tomorrow.

Download & install

Clone, install, run. Requires Node.js ≥ 22.5 and Git — no native modules, no build tools.

PowerShell — install Node LTS (≥ 22.5) via winget, then Shiba Studio:
# prerequisites
winget install OpenJS.NodeJS.LTS
winget install Git.Git

# install & run Shiba Studio
git clone https://github.com/stevologic/shiba-studio.git
cd shiba-studio
npm install
npm run dev
Open http://localhost:3000 → Settings → connect your xAI key. Production mode: npm run build && npm run start.
Terminal — install Node (≥ 22.5) via Homebrew, then Shiba Studio:
# prerequisites
brew install node git

# install & run Shiba Studio
git clone https://github.com/stevologic/shiba-studio.git
cd shiba-studio
npm install
npm run dev
Open http://localhost:3000 → Settings → connect your xAI key. Production mode: npm run build && npm run start.
Shell — install Node (≥ 22.5) via nvm (works on every distro), then Shiba Studio:
# prerequisites
curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.1/install.sh | bash
nvm install 22

# install & run Shiba Studio
git clone https://github.com/stevologic/shiba-studio.git
cd shiba-studio
npm install
npm run dev
Open http://localhost:3000 → Settings → connect your xAI key. Production mode: npm run build && npm run start.