Artificial intelligence is moving fast, but for many organizations there’s still an uncomfortable reality. Most AI tools today require sending data to somebody else’s cloud. That can make companies uneasy, especially when sensitive documents, internal research, or proprietary information are involved.
Now Mozilla’s email-focused subsidiary, MZLA Technologies, is trying to tackle that concern with a new project called Thunderbolt. The company is positioning it as an open-source AI client designed for organizations that want to run AI on their own terms.
Instead of forcing businesses into a particular AI ecosystem, Thunderbolt is meant to be flexible. Organizations can host it themselves, connect it to their own data sources, and decide which models power the system. That could mean commercial AI services, open-source models, or even fully local deployments depending on what a team prefers.
The idea is to turn AI into something that feels less like a standalone tool and more like a workspace. Users can interact with models through chat, search, and research tools while connecting the system to internal databases and enterprise software. From there, organizations can build automated workflows, generate reports, monitor topics, or trigger actions based on events and schedules.
Thunderbolt is also designed to tie into a broader AI infrastructure layer through integration with Haystack, an open-source framework from the AI company deepset. Haystack is often used for building agent systems and retrieval augmented generation pipelines, which help AI models work with large collections of data.
When paired together, the Thunderbolt interface and the Haystack backend create what Mozilla describes as a unified AI stack. In theory, that means the user experience and the infrastructure behind it are closely connected rather than existing as separate tools scattered across an organization.
For businesses that want control over where AI runs, this approach could be appealing. A self-hosted client means internal data stays inside company infrastructure instead of being routed through outside services. Mozilla also says organizations can add optional end to end encryption and device level access controls if they want tighter security.
Another practical detail is platform support. Thunderbolt is intended to run across Windows, macOS, Linux, iOS, and Android, which should make it easier for teams to use the same AI workspace across different devices.
Conceptually, the project makes sense. Plenty of organizations are excited about AI but remain cautious about relying entirely on third party platforms. A self-hosted option that still offers modern AI capabilities could fill a gap that many companies feel today.
That said, I do have one gripe. The name Thunderbolt is not great. There are already countless products and technologies called Thunderbolt in the tech world. Most folks immediately think of the hardware interface used on laptops and desktops. Launching another product with the same name feels confusing and, frankly, a bit lazy. Mozilla probably could have chosen something more distinctive.
Still, the broader idea here is interesting. As AI becomes more embedded in daily work, open-source alternatives that give organizations independence may start to attract more attention.
Thunderbolt is available now through a waitlist, and the source code has been published publicly. Enterprise deployments will vary depending on support and customization needs, but the core concept is clear. Mozilla wants organizations to have the option to run AI without handing over the keys to someone else’s platform.
