Skip to content
robsccPublic

About

Local-first personal AI agent desktop app — Claude Code's shape with a visual workflow studio, local knowledge base, and zero cloud. Tauri + Next.js + Python.

Topics

Resources

Contributing

Security policy

Stars

14 stars

Watchers

0 watching

Forks

Repository files navigation

Ginno

A local-first personal AI agent that runs on your own machine.

The shape of Claude Code, all-file storage (no DB · no account · no cloud), and a visual Workflow Studio.

License: MIT CI codecov Platform No cloud

English · 简体中文

Ginno chat workspace


What it is

Ginno turns the capabilities an AI agent should have into a complete desktop app, instead of scattering them across a terminal and a dozen browser tabs.

  • 🧠 Multi-agent chat — switch agents mid-conversation, shared history, per-agent memory
  • 🛠️ Tools · permissions · skills · MCP · hooks — Claude Code's shape, all built in
  • 🔀 Visual Workflow Studio — a versioned DSL you compile into a LangGraph state graph, with a live run observer
  • 📚 Local knowledge base — index your Obsidian vault; answers come back with citations
  • 🎯 Long-horizon goals + TODO sync — let the agent keep working toward a standing objective
  • 🌐 Embedded browser + web search — retrieval with source attribution, not vibes

Full feature walkthrough: docs/user-guide.md.

Workflow Studio Local knowledge base
Workflow Studio Knowledge base

Why "local-first"

One design rule: your data should not leave your machine.

Ginno
Database ❌ none — all state is JSON / JSONL / Markdown under ~/.ginno/
Account ❌ none — open it and go
Cloud sync ❌ none — everything stays local
Portable ✅ copy the folder to move it; readable, hand-editable, git-able

You bring your own model key (OpenAI / Anthropic / DeepSeek, or any compatible endpoint). Ginno is just a client — it never takes custody of your key or your data.


Install

Option A — download (macOS, Apple Silicon): grab the latest .dmg from Releases (built automatically by the release pipeline). The build is ad-hoc signed, not notarized — on first launch use right-click → Open to get past Gatekeeper. See the upgrade guide when moving between versions.

Option B — build from source:

git clone https://github.com/robscc/ginno
cd ginno

pnpm install                       # web + desktop shell
cd packages/runtime && uv sync     # Python runtime
cd ../..

pnpm dev                           # web + runtime + desktop shell, all at once

Requires: Node ≥ 20 · pnpm ≥ 9 · Python ≥ 3.11 (via uv) · Rust (Tauri).

Package a .app: make app → apps/desktop/target/release/bundle/dmg/Ginno_*.dmg (see docs/p3-packaging-notes.md).


Stack

Layer Choice
Shell Tauri 2 (Rust) — process management + webview only
UI Next.js 14 (static export) + React 18 + Tailwind
Runtime Python + FastAPI + LangGraph (sidecar, bundled with PyInstaller)
Storage plain files (no DB); optional LanceDB for semantic-search vectors

Architecture and subsystem designs: docs/architecture.md.


Documentation

User guide full walkthrough of every feature
FAQ · Troubleshooting common questions · known failure modes & fixes
Upgrade guide moving between versions — your ~/.ginno data survives
How it compares vs Claude Code / Cline / Open WebUI / LangGraph Studio
ROADMAP · CHANGELOG where it's heading · what shipped
CONTRIBUTING · SECURITY dev setup, testing, release runbook · vulnerability disclosure

Status

  • Every push runs a full-stack CI — web (lint/typecheck/vitest/build), Rust (fmt/check), runtime (2900+ tests with coverage) — all green.
  • A personal project, used daily (dogfooding) and iterated fast — ~3.5 months, 490+ commits.
  • Early and unpolished; APIs and UI still move.
  • Issues, PRs, and ideas are welcome — in English or Chinese (see CONTRIBUTING).

License

MIT © 2026 ChuanchuanSong

About

Local-first personal AI agent desktop app — Claude Code's shape with a visual workflow studio, local knowledge base, and zero cloud. Tauri + Next.js + Python.

Topics

Resources

Contributing

Security policy

Stars

14 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages