Pinvou/pinvou-agent

What it solves

Pinvou Agent provides a unified desktop workspace that transforms AI chat into a result-oriented environment. It solves the problem of fragmented AI interactions by integrating work, visual design, and software development into a single interface where the goal is to produce a final deliverable (like a file or a visual artifact) rather than just a text response.

How it works

The application is built using Tauri 2, React, and Vite, with a core agent engine called CodeWhale. It connects to local vLLM instances or any OpenAI-compatible API. The system uses a multi-session workspace where messages, tool calls, and artifacts persist. It integrates MCP (Model Context Protocol) servers, CLI connectors, and a local knowledge base with vector retrieval to provide the agent with real-world tools and context.

Who it’s for

It is designed for professionals in work, design, and coding who want a private, extensible AI agent workspace on their desktop that can handle complex tasks, manage files, and integrate with existing software tools.

Highlights

• Three Specialized Modes: Dedicated workflows for general work (research/writing), visual design (prompt-to-editable-visual), and coding (integrating agents like Claude Code or Codex into real projects).
• Artifact Panel: A centralized area that collects and previews every file the agent creates or edits.
• Extensible Tooling: Support for local and remote MCP servers, CLI tools, and pre-made connectors for platforms like Obsidian, Feishu, and DingTalk.
• Local Knowledge Base: Full-text and vector retrieval for personal files with provenance tracking in answers.
• Privacy-First: Ability to run fully local loops using local models and tools.