The autonomous AI agent
for your desktop

ORCA plans, writes code, runs it, reads the errors and fixes them — on your machine, in your files. Built-in models work out of the box; bring your own keys for 200+ providers or any OpenAI-compatible endpoint.

Download ORCA Read the docs GitHub

Windows 10/11 · macOS 12+ (Apple Silicon & Intel) · Linux x64 — self-updating, SHA-256 verified, MIT licensed.

46 real tools

Shell, Node/Python, precise file editing, web search, HTTP, headless-browser checks, Office documents, media, OCR, image & video generation, background processes, sub-agents.

Any model

Built-in models with automatic failover, plus your own cloud or local providers (Ollama, LM Studio, vLLM). Keys never leave your device.

MCP client

Connect any Model Context Protocol server over stdio or Streamable HTTP — presets included, per-tool toggles, untrusted-content isolation.

Builds whole sites

The model writes the content; scaffold_site renders a finished multi-page site and every page is verified in a headless browser.

Checkpoints & diffs

Every file change is snapshotted; restore any step with one click. Live edits render as reviewable diff cards.

Private by design

No accounts, no telemetry. Everything lives in a local data folder you control, behind an authenticated local API.

Quick start

bash

git clone https://github.com/Nethyric/orca.git
cd orca
npm install
npm start          # desktop app
npm run web        # or: browser UI at http://localhost:7860

Prefer a package? Grab the build for your system from Releases — the app updates itself from there, verified against SHA256SUMS.

The autonomous AI agentfor your desktop