Thunderbolt is an open-source, cross-platform AI client that you can deploy on-premises anywhere.
The tool addresses vendor lock-in and data ownership concerns by letting you choose which AI models to use, whether frontier models via API, local models through Ollama or llama.cpp, or on-premises deployments. It runs on web, iOS, Android, Mac, Linux, and Windows, with a backend you can self-host using Docker Compose or Kubernetes. You add your own model providers and API keys in settings, maintaining full control over which services process your data.
Thunderbolt targets enterprise customers planning on-premises deployment who want to avoid dependence on a single AI vendor. It suits organizations with security or compliance requirements that demand local or private infrastructure. The project is early-stage and still undergoing security audit, so it is not yet production-ready for general use. Current limitations include a dependency on authentication and search functionality, though search can be disabled; the team plans eventual full offline-first operation. If you need a mature, battle-tested AI client with minimal setup friction, this is not yet the right choice. If you are willing to self-host and configure your own models, and you prioritize data sovereignty over convenience, it warrants evaluation.
The project maintains active development with ongoing work toward enterprise production readiness. Documentation covers deployment, development setup, system architecture, and release processes. The team welcomes contributions and has established community participation guidelines. Development infrastructure includes tooling for component testing via Storybook, bundle analysis, and signing key management for releases.
