A local AI chat client for OpenAI and OpenAI-compatible endpoints.
Open it in your browser: ai.dev-team.org
AI Client is a desktop and web application for working with AI models on your own computer. It speaks the OpenAI Responses API and the Chat Completions fallback, so it works with OpenAI, Azure, local models, and any other provider that follows the same protocol.
Everything runs locally. A small Host process stores your projects, chat history, and credentials, and the Blazor WebAssembly UI renders in your browser or in the desktop shell. Tools reach the model through MCP (Model Context Protocol), so the assistant can read, edit, search, and run commands in directories you grant.
- Streaming chat with Markdown — see the response as it is generated, with rendered code blocks, tables, and safe links.
- Git-like branching — fork any message into a new branch to try a different path without losing the original.
- Local MCP tools — give the model access to files, search, and other tools, with per-project policies and confirmation prompts.
- Multiple providers — connect to several endpoints and pick a model per chat. Credentials stay on your machine.
- Projects with isolation — each project has its own endpoint, MCP servers, file system roots, and system prompt.
- Cross-platform — Chrome or Edge on any OS, a native desktop app for Windows, macOS, and Linux, or install the web app from your browser.
- Open ai.dev-team.org in Chrome or Edge.
- Download the local Host for your platform, or pick a desktop installer for Windows, macOS, or Linux.
- Create a project, add your endpoint and credential, allow a directory, and start chatting.
The Host is a small background process that holds your data. Close it when you are done; your projects and history stay in the local data directory.
C# scripting tools (AI.Mcp.CSharp) are optional. Select C# scripting tools in the Windows or macOS Host/Desktop installer. On Linux, install the companion AI.Mcp.CSharp-<runtime>.deb package alongside Host or Desktop, in either order. The scripting server includes its own .NET runtime.
Build the installers on the matching operating system with dotnet run --project build -- package-release --runtime win-x64 --version 1.0.0 (also supports win-arm64, osx-x64, osx-arm64, linux-x64, and linux-arm64). Output is in artifacts/release/packages; Windows packaging requires Inno Setup 6, macOS uses pkgbuild/productbuild, and Linux uses dpkg-deb.
Updates are managed in Settings → Updates. Desktop updates Desktop; the browser updates its local Host. The Preview channel includes preliminary releases and is the initial default. Both products check at startup and every six hours and download updates automatically. Host also installs automatically after tasks finish; Desktop requires enabling auto-installation or selecting Restart and update. Drafts and workspace navigation survive the restart. Installation retains the optional C# scripting tools and verifies the package SHA-256 supplied by GitHub.
macOS and Linux may request administrator authorization. Linux uses pkexec, or sudo with zenity for the password prompt; background Host updates also require the user systemd session. Developer builds support manual checks but do not install updates. Older installations without update support need one manual upgrade first.
The full guide, including project structure, MCP setup, and the security model, lives in docs.
AI Client is licensed under the Apache License, Version 2.0. See NOTICE for attributions.