Who Thunderbolt is for#
Enterprise teams evaluating on-prem AI clients
Use Thunderbolt when the priority is controlling data flow and model routing rather than adopting a consumer-hosted AI app.
Skip if:
You need a fully managed AI assistant with no backend or provider setup.
Teams comparing local and hosted models
Use Thunderbolt when developers need to switch between Ollama, llama.cpp, and OpenAI-compatible providers from one client.
Skip if:
Your organization has already standardized on one vendor app and does not allow alternative clients.
Security teams testing AI workflows
Use Thunderbolt to evaluate on-prem AI UX patterns before rolling them out broadly.
Skip if:
You require a mature production certification package today.
The problem it solves#
AI clients often bundle the interface, account system, hosted inference, and data policy into one vendor relationship. That creates lock-in when model quality, pricing, or compliance needs change. Enterprises also need clearer control over where prompts and files are processed, especially when internal data cannot be sent to a consumer AI product by default.
How it solves it#
Cross-platform AI client
Web, iOS, Android, macOS, Linux, and Windows support gives teams one client direction across mixed device fleets.
Bring your own model provider
Thunderbolt can work with local inference through Ollama or llama.cpp and with OpenAI-compatible providers. Teams can choose models without replacing the client experience.
On-prem deployment path
The project targets enterprise customers that want to deploy on-prem and points to Docker Compose or Kubernetes deployment documentation.
Local development stack
`make doctor`, `make setup`, `make up`, and `make run` bring up the backend, frontend, Postgres, and PowerSync locally.
Strengths and trade-offs#
Strengths
- Model choice is centralThunderbolt separates the client from the model provider decision. This matters when teams want to compare hosted APIs, local models, or on-prem inference without changing daily user workflows.
- Clear enterprise control storyThe project explicitly targets on-prem deployment and data ownership. That makes it more relevant to security teams than consumer AI chat clients.
- MPL-2.0 licenseThunderbolt is licensed under MPL-2.0, giving teams source access while preserving file-level copyleft obligations.
Trade-offs
- -Still under active developmentThunderbolt is early, undergoing security audit work, and preparing for enterprise readiness. Treat it as an evaluation candidate, not a drop-in mature product.
- -No public inference endpointUsers must add their own model providers. That is useful for control, but less convenient than ChatGPT or Claude Desktop where hosted inference works immediately.
Install and self-host#
make doctor
make setup
make up
make runWhat it's built on#
- Languages
- TypeScript
- Frameworks
- React
FAQ#
Is Thunderbolt production-ready?
Not fully. Thunderbolt is early, under active development, undergoing a security audit, and preparing for enterprise production readiness.
Can Thunderbolt use local models?
Yes. Thunderbolt supports Ollama and llama.cpp for local inference, plus OpenAI-compatible providers.
Can Thunderbolt be self-hosted?
Yes. Thunderbolt provides Docker Compose and Kubernetes deployment paths for self-hosting.
Similar open-source tools#
VibePad
Control your AI coding assistant with a gamepad, open source
orca
The ultimate IDE for coding agents
agentmemory
Persistent memory for AI coding agents
Agent Skills
Structured workflows for AI coding agents.
Flue Framework
Build powerful, autonomous agents with TypeScript.
jcode
Next-gen coding agent harness for efficient workflows