# Mozilla's Thunderbolt: Open-Source AI Client for Enterprise Control Over AI
The Mozilla Foundation has launched Thunderbolt—an open-source client for working with AI, designed for deployment on your own infrastructure. Unlike commercial solutions like Copilot or ChatGPT Enterprise, Thunderbolt lets organizations maintain full control over their data and avoid dependency on a single AI service provider.
Architecture and Deployment Model
Thunderbolt was developed by Mozilla's subsidiary MZLA Technologies, known for the Thunderbird email client. The project code is released under the MPL 2.0 license and available on GitHub. It supports both a web version and native apps for all major platforms: Windows, macOS, Linux, iOS, and Android.
The core architectural principle is complete decentralization of data processing. The client interacts with AI models through a proxy layer that can be deployed locally or in a private cloud. This allows seamless integration with public models (OpenAI, Anthropic, Mistral) as well as local solutions via Ollama or llama.cpp, and even internal corporate LLMs.
Integration with business systems is handled through:
- The Haystack platform from deepset (Berlin)
- MCP servers (Model Context Protocol)
- Agents via the ACP protocol (Agent Client Protocol)
This approach ensures flexibility when connecting to existing IT landscapes without needing to rewrite internal services.
Usage Modes and Features
The Thunderbolt interface is organized around the concept of "workspaces" with several usage modes:
- Chat — classic conversational interaction with the model.
- Search — semantic search across internal documents and knowledge bases.
- Research — analysis of large volumes of text with automatic structuring of outputs.
- Scheduled Tasks — automation of routine operations, such as:
- Daily briefings
- Monitoring topic trends
- Scheduled report generation
All these features run on a unified architecture where data never leaves the trusted zone unless explicitly allowed by the organization's security policy.
Security and Compliance
Security in Thunderbolt rests on three pillars:
- Self-hosting by default: data stays within the client's infrastructure.
- Optional end-to-end encryption: even with external models, content can be encrypted before transmission.
- Device-level access control: admins can restrict AI usage to specific devices or roles.
The project is currently undergoing an independent security audit. While it's not yet officially recommended for production deployment, Mozilla is actively testing it with partner companies.
Business Model and Roadmap
Despite its open-source core, MZLA Technologies plans to monetize through two channels:
- Managed hosting: a cloud version for teams who prefer not to handle self-hosting.
- Professional support: SLA guarantees, custom integration, and training for enterprise clients.
This balances the open-source philosophy with sustainable funding for ongoing development.
Key Takeaways
- Thunderbolt is more than just a chat interface—it's a full-fledged enterprise AI platform supporting multiple model providers.
- All data stays within the client's infrastructure by default, which is critical for regulated industries.
- Integration with Haystack, MCP, and ACP makes it compatible with modern enterprise stacks.
- The project is in pre-release but ready for testing in controlled environments.
- Mozilla positions the product as a counter to AI market monopolization by OpenAI and Microsoft.
— Editorial Team
No comments yet.