Witsy is a desktop AI assistant and universal MCP client built with TypeScript, offering local HTTP API endpoints, an integrated CLI, and multi-provider support for chat, embeddings, and various model providers. It ships with multiple releases since 2026 and provides a BYOK model setup and local/remote usage options.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
Witsy is a desktop AI assistant / universal MCP client built with TypeScript. It functions as a BYOK AI application that can connect to various large language models (OpenAI, Anthropic, Google Gemini, MistralAI, Ollama, and others) and supports MCP servers, embeddings, and multiple media capabilities. It includes a local HTTP API, a CLI, and a setup process that allows binary downloads or self-build options.
How it works
Witsy exposes a local HTTP API (default port 8090) with endpoints for health, chat, scratchpad, settings, studio, forge, realtime, prompt, command, transcription, readaloud, engines, models, complete, agent webhook/status, and more. It can run MCP servers with various LLMs and supports embedding/document search (RAG) via local repositories. The README details endpoints and example curl usages, plus a command-line interface that connects to the running app and provides commands to manage models, port, history, and exit.
Getting started
Prerequisites include API keys for supported providers and Ollama for local model runs. Setup options include downloading a binary from releases or building locally with npm. The basic setup steps shown are:
npm ci
npm start
The README also lists API-key prerequisites for OpenAI, Anthropic, Google, xAI, Meta, MistralAI, DeepSeek, OpenRouter, Groq, Cerebras, and Ollama models, plus TTS and STT providers.
There is a CLI that gets installed when launching Witsy for the first time. The CLI installation details indicate symlinks or PATH adjustments on macOS/Windows/Linux, and usage begins with witsy once the server is running.
Recent releases
Latest five releases are:
- v3.5.2 3.5.2 (2026-03-04)
- v3.5.1 3.5.1 (2026-02-21)
- v3.5.0 3.5.0 (2026-01-30)
- v3.4.2 3.4.2 (2026-01-23)
- v3.4.1 3.4.1 (2026-01-19)
Each release links to CHANGELOG.md for details.
Traction
The repository reports 2012 stars and 166 forks, with 55 open issues as of the provided data. These numbers come directly from the repo metadata and are not interpreted.
Behind the repo
The repository lists topics indicating integrations and providers such as anthropic, deepseek, ollama, openai, googleai, mistral, and various MCP-related terms. There is no separate startup/company profile section in the provided facts beyond these topics.
Caveats
- License: AGPL-3.0
- Created: 2024-04-25
- Last push: 2026-04-23
- Open issues: 55
- Language: TypeScript
The README includes prerequisite API keys for multiple providers and notes that OpenAI/Ollama usage may require local model files or API keys, but does not add license or distribution caveats beyond the AGPL-3.0 listing.