Overview
Thunderbolt is an open-source AI client from the Thunderbird project, built around one idea: you choose the models and you keep the data. It runs on every major desktop and mobile platform — web, macOS, Linux, Windows, iOS and Android — and it is designed to be deployed on-premises anywhere rather than consumed as someone else's hosted service.
There is no bundled inference endpoint. You point Thunderbolt at the providers you already have: Ollama or llama.cpp for free local inference, or an API key for any OpenAI-compatible provider entered in the app's settings. That keeps frontier, local and on-prem models behind the same interface, which is the vendor-lock-in problem the project is explicitly aimed at.
The project is early and says so up front. It is under active development, undergoing a security audit, and currently targets enterprise customers who want an on-prem deployment; the team's stated plan is to become fully offline-first, but today the app still depends on authentication and on search (which can be disabled from the integrations screen). The stack is a TypeScript frontend with a Tauri shell, a backend, Postgres and PowerSync, all licensed MPL 2.0.
What it does
- Runs on web, iOS, Android, macOS, Linux and Windows from one codebase
- Bring your own models — frontier APIs, on-prem endpoints, or local inference through Ollama or llama.cpp
- Any OpenAI-compatible provider can be added with an API key from the settings screen
- Designed for on-prem deployment, with Docker Compose and Kubernetes paths documented under deploy/
- MPL 2.0 licensed, with published telemetry and privacy documentation
- Enterprise features, support and forward-deployed engineers offered by the Thunderbird team
Getting started
The repo ships a Makefile that walks you through a local development environment. Run `make doctor` first — it checks your toolchain and prints the exact install command for anything missing. For a real deployment, follow deploy/README.md instead.
Check your toolchain
make doctor verifies the tools Thunderbolt needs and prints install commands for anything you are missing.
make doctorInstall dependencies
Installs frontend and backend dependencies and wires up the agent symlinks.
make setupStart the services
Brings up Postgres and PowerSync in Docker, then runs the backend on :8000 and the frontend on :1420.
make up
make runAdd a model provider
Thunderbolt ships no public inference endpoint. Add your own provider in the app settings — Ollama or llama.cpp for local inference, or an API key for any OpenAI-compatible provider. Search can be turned off on the integrations screen if you want less network dependency.
Self-host it properly
For a real deployment, follow deploy/README.md, which covers Docker Compose and Kubernetes. You deploy your own backend and sign up against it.
Commands and code are distilled from the project's own documentation — always check the official repo for the latest.
When to use it
- Give an organisation a single AI client across desktop, mobile and web while keeping inference and data inside its own infrastructure
- Run a private chat client entirely against local models via Ollama or llama.cpp
- Avoid vendor lock-in by putting frontier APIs, on-prem endpoints and local models behind one interface you can switch between
- Evaluate an on-prem AI client in a regulated environment where a hosted assistant is not an option
How Thunderbolt compares
Thunderbolt alongside other open-source chat uis tools AI/TLDR tracks, ranked by GitHub stars.
| Tool | Stars | What it does |
|---|---|---|
| Open WebUI | ★ 153k | A self-hosted web interface for chatting with local and API-based LLMs (Ollama, OpenAI-compatible) that adds RAG, web search, image generation, and multi-user access. |
| Lobe Chat | ★ 82.7k | An open-source chat UI with a plugin and agent marketplace, multi-model support, voice, vision, and one-click self-hosting. |
| AnythingLLM | ★ 66.2k | A desktop and self-hosted chat app focused on document Q&A and agents, letting you build private RAG workspaces over your own files and data sources. |
| Cherry Studio | ★ 52k | A cross-platform desktop chat client that connects to many cloud and local model providers and ships with hundreds of preset assistants and MCP support. |
| LibreChat | ★ 44.4k | A self-hostable ChatGPT-style chat app that unifies many AI providers and adds agents, MCP support, artifacts, a code interpreter, and multi-user authentication. |
| Chatbox | ★ 41.8k | A desktop and web chat client for ChatGPT, Claude, Gemini, Ollama, and other models, with local message storage and multi-provider switching. |
| Buzz | ★ 33.7k | A self-hostable team workspace built on a Nostr relay where AI agents join channels as members with their own keys, and messages, patches, workflow runs and approvals are all signed events in one log. |
| Thunderbolt | ★ 4.8k | Cross-platform AI client you can deploy on-prem, with your own model providers |