Pinvou/pinvou-agent

Open-source desktop AI agent for tools, files, knowledge, workflows, and real deliverables.

Pinvou Agent – Desktop AI‑assistant workspace

What it is – An open‑source desktop application that lets you work with a conversational AI as a real‑world assistant. It combines chat, file handling, knowledge‑base search, and tool integration into three “modes”: Work (research, writing, analysis), Design (turn prompts into editable posters or charts) and Code (bring a coding‑agent into an actual codebase). The app can run entirely locally (e.g., with a vLLM server) or connect to any OpenAI‑compatible API, and it is extensible via plug‑in‑style Skills, MCP servers, CLI tools, and third‑party connectors.

Core concepts & workflow

Concept What you get
Workspace A titled session that persists chat history, tool calls, and generated artifacts (files, markdown, images, etc.).
Artifacts panel Automatic collection of every file the agent creates/edits; you can preview, open, or further edit them.
Plan / YOLO modes Choose to see a step‑by‑step plan before execution, or let the agent run straight‑through when the task is clear.
Knowledge base Local vector/full‑text search over attached PDFs, Office docs, images, etc.; you can enable/disable collections per chat.
Persona cards Saved specialist roles (e.g., “legal analyst”, “designer”) that bias the model’s behavior.
Skills & Workflows Reusable command bundles that the agent can invoke as stable, repeatable capabilities.
Tool store Central catalogue for local MCP servers, remote MCP servers, CLI tools, and API connectors (Feishu, DingTalk, Obsidian, etc.).
Remote control Scan a QR code to view/steer a running workspace from a phone.
Voice input Optional on‑demand speech‑recognition models for hands‑free prompting.

Notable features

  • Multi‑session UI with title search, so you can jump between projects.
  • Attachment handling – drag‑and‑drop PDFs, Office files, images, archives; the agent can read and act on them.
  • Editable markdown artifacts – edit directly or ask the agent to rewrite selected passages.
  • Local knowledge base with vector retrieval, keeping provenance of each file.
  • Memory center that stores long‑term preferences and asks you to confirm candidate memories.
  • Extensible connector ecosystem – OAuth/SSO where supported, plus ready‑made connectors for popular Chinese enterprise tools.
  • Coding mode supports Codex, Claude Code, or Kimi via the ACP (agent‑code‑protocol) allowing the agent to read/write real project files, run commands, and show a plan of actions.
  • Design mode lets you generate posters or data visualisations from natural‑language prompts and then edit them graphically within the app.

Tech stack

  • Tauri 2 – native desktop wrapper (Rust backend, WebView front‑end). Provides cross‑platform builds for Windows, macOS, Linux.
  • React 19 + Vite – UI layer.
  • Rust engine (CodeWhale submodule) – handles model calls, streaming, tool execution, MCP server orchestration, Skills, and session management.
  • Model access – works with any OpenAI‑compatible endpoint (OpenAI, Anthropic, Gemini, DeepSeek, Kimi, Qwen, etc.) and with local vLLM servers.
  • File‑system layout – all user data lives under ~/.pinvou3/ (sessions, knowledge collections, settings, extensions).

Getting started (quick‑start)

  1. Install prerequisites: Git (with submodule support), Node ≥ 18, a recent Rust toolchain, and Tauri 2 system dependencies for your OS.
  2. Clone the repo recursively:
    git clone --recursive https://github.com/Pinvou/pinvou-agent.git
cd pinvou-agent/pinvou3-app
npm ci
cd ..
./pinvou3-app/run-dev.sh
  3. Add a model configuration in the app settings (local vLLM URL or cloud API key).
  4. Create a new workspace, attach files, pick a persona or skill, and start a task in Work, Design, or Code mode.

Pre‑built binaries are also available on the Releases page for all three platforms.

Extensibility

  • Skills – add a SKILL.md package describing a tool or method; the agent can invoke it from any session.
  • MCP servers / CLI connectors – plug in external services (e.g., enterprise knowledge bases) via the unified tool store.
  • Custom instructions – bundle instructions.md with a skill to steer model behaviour.
  • UI changes – modify the pinvou3-app/ folder; engine changes go to the CodeWhale/ submodule.

Community & governance

  • Chat & support – QQ group (Chinese) #1108909346, GitHub Issues, Discussions.
  • Security – private reporting channel defined in SECURITY.md.
  • Contribution – follow CONTRIBUTING.md and the fork‑policy for the CodeWhale engine.

License

MIT – permissive open‑source license, with third‑party attributions listed in THIRD_PARTY_NOTICES.md.

Bottom line – Pinvou Agent is a full‑featured, cross‑platform desktop environment that turns a conversational LLM into a practical assistant for everyday work, visual design, and software development, while keeping data private when you run it locally. It is built on modern web‑native (React/Vite) and native (Tauri/Rust) technologies and is designed to be extended through a modular skill/connector ecosystem.

Related

  • Project
    hhyqhh/inno-agent

    Inno Agent is an open‑source personal learning AI built on the Pi coding‑agent SDK. It adds three memory layers (learner profile, wiki, session logs), a natural‑language scheduler, IM channel support (Feishu/WeChat), and an in‑browser practice lab. Available as an Electron desktop app, a React web UI, or a terminal CLI, it works with any OpenAI‑ or Anthropic‑compatible model and can be extended via a skill hub. The project is MIT‑licensed and includes Docker, sandboxing, and a session‑replay showcase.

  • Project
    iOfficeAI/AionUi

    AionUi is an open‑source, cross‑platform desktop app that turns LLM APIs into autonomous office assistants. It ships with a built‑in Rust agent (zero‑setup), supports 30+ AI providers, can read/write Word/PPT/Excel files via OfficeCLI, integrates dozens of existing CLI agents, offers team‑mode orchestration, scheduled cron jobs, and remote access through a WebUI or chat‑bot bridges (Telegram, WeChat, etc.). Ideal for developers, analysts, and power‑users who want AI to act on their files and automate tasks without writing glue code.

  • Project
    RizRiyz/luvus

    Luvus is a cross‑platform terminal UI that lets you launch, monitor, and orchestrate AI coding agents (Claude, Codex, Grok, etc.) alongside shells, editors, and Git tools. It provides persistent workspaces, pane management, agent‑aware status panels, Git/GitHub integration, remote‑machine support, a read‑only web view, and an extensible module system. Install via a script, Homebrew, or PowerShell, then use `luvus` to start a session, `luvus doctor` to verify setup, and `luvus update` for upgrades.

  • Project
    OpenCoworkAI/open-cowork

    Open Cowork is a free, open‑source desktop app that wraps Claude‑style (and other) LLMs into a local AI‑agent workspace. With one‑click installers for Windows/macOS, VM‑level sandboxing (WSL2/Lima), a built‑in “Skills” system for generating PPTX/DOCX/PDF/XLSX files, MCP connectors for browsers/Notion, and remote control via Feishu or Slack, it lets non‑technical users give natural‑language commands that read/write files, automate GUI apps, and produce professional documents—all running locally with optional VM isolation.

  • Project
    smallnest/pigo

    pigo is a Go‑based CLI AI coding assistant (a clone of the pi agent) that can read/write files, run shell commands, fetch web content, and interact with many LLM providers. It works in headless script mode or an interactive REPL, supports tool‑level allow/deny lists, project trust, prompt templates, skills, plugins, self‑update, and provides a Go SDK for embedding.