For the past couple of years, injecting artificial intelligence into a business workflow has usually meant handing over the keys to a tech giant. Tools like Microsoft Copilot, Claude Enterprise, and ChatGPT Enterprise are undeniably powerful, but they often force organisations to upload their highly sensitive, proprietary data to third-party cloud servers. Mozilla is officially pushing back against this standard. Through its MZLA Technologies subsidiary—the same team behind the popular Thunderbird email client—the company has announced Thunderbolt, an open-source AI workspace designed to give businesses total control over their digital infrastructure.
The Push for Sovereign AI
Mozilla is framing Thunderbolt as a “sovereign AI client”. The core philosophy driving the project is straightforward: choose your own models, own your own data, and permanently eliminate vendor lock-in. Instead of acting as another massive, closed-box subscription service, Thunderbolt operates as an extensible user interface that allows organisations to securely host and govern their own AI environments.
If an enterprise is uncomfortable sending financial records or proprietary code across the open web, Thunderbolt allows them to process everything internally. The software can be scaled down to run securely on a single local machine or deployed widely across a massive internal corporate network.
A Flexible AI Architecture
To make this high level of autonomy work, Mozilla has partnered heavily with the German firm Deepset to integrate its Haystack platform directly into the Thunderbolt experience. This acts as an open-source orchestration layer, allowing developers to easily build custom retrieval-augmented generation pipelines and connect their AI agents to massive pools of local enterprise data. The client also fully supports the Model Context Protocol and Agent Client Protocol.
Crucially, Mozilla is not actually providing the brain power here; the company does not offer a public inference endpoint. Instead, Thunderbolt acts as a universal shell. Users are free to plug in major commercial application programming interfaces from OpenAI or Anthropic, or they can opt for completely free, offline processing by running local models through tools like Ollama or llama.cpp.
Enterprise Privacy and Everyday Usability
For end users, the experience is designed to feel as polished as any proprietary tool. Employees can use Thunderbolt to generate daily briefings, summarise local documents, automate scheduled tasks, conduct deep research, and chat with customised AI agents tailored to their specific departmental needs. To ensure the platform fits seamlessly into modern hybrid work environments, Mozilla is not limiting the software to a single ecosystem. Native applications are rolling out for Windows, macOS, Linux, iOS, and Android, alongside a standard web application, ensuring employees can securely access their corporate AI from essentially any hardware device.
On the backend, security remains the primary selling point. Because the platform utilises an offline SQLite database as its local knowledge repository, the entire software stack remains under the company’s direct control. Furthermore, the system includes optional end-to-end encryption and strict device-level access controls to ensure sensitive queries remain private.
Availability and Deployment
While the software is incredibly promising, Mozilla is clear that Thunderbolt is currently in its early stages and is actively undergoing a formal security audit. However, the core React source code is already available publicly on GitHub under the Mozilla Public License 2.0 for developers who want to start experimenting immediately. For smaller teams or enterprises that want the benefits of an open-source workspace but lack the technical expertise to set up the infrastructure themselves, MZLA Technologies is currently taking signups to provide hosted, guided deployments and paid enterprise licensing in the near future.
