Thunderbolt from Mozilla: open AI client for business

# Mozilla's Thunderbolt: Open AI Client for Enterprise Infrastructure

Mozilla has introduced Thunderbolt — an open-source AI client designed for organizations and developers seeking full control over their AI infrastructure. Unlike cloud solutions with proprietary architectures, Thunderbolt lets you deploy a workspace locally, integrate your own data, and select models from a wide range of sources — from commercial providers to open-source options.

Architecture and Key Features

Thunderbolt is positioned as a "sovereign" AI client, giving users complete control over data, models, and compute resources. The architecture is built around principles of extensibility and compatibility with existing standards. The client supports connections to:

• the Haystack platform from deepset;
• Model Context Protocol (MCP) servers;
• agents implementing Agent Client Protocol (ACP).

This provides flexibility for integration into enterprise environments where specialized AI pipelines or internal LLM services are already in use.

Model Support and Local Deployment

One of Thunderbolt's core advantages is its independence from specific AI providers. Users can connect:

• Commercial models via API (e.g., OpenAI, Anthropic);
• Open models such as Llama 3, Mistral, or Qwen;
• Locally deployed models via Ollama, llama.cpp, or custom inference servers.

All interactions with models occur within the client's secure environment. With local deployment, data never leaves the organization's perimeter — a critical feature for regulated industries like finance, healthcare, and government.

Automation and Cross-Platform Support

Thunderbolt includes built-in automation tools. Users can configure:

• Daily analytics summaries;
• Real-time topic monitoring;
• Scheduled report generation;
• Event-driven responses (e.g., notifications on system data changes).

The client is available as a web app and native builds for all major OSes: Windows, macOS, Linux, iOS, and Android. This enables seamless transitions between devices without losing session context.

Security and Licensing

Security is implemented on multiple levels:

• Full local deployment capability (on-premises);
• Optional end-to-end encryption;
• Device- and user-level access management.

The project code is distributed under the MPL 2.0 license and available on GitHub. For enterprise customers, Mozilla MZLA Technologies Corporation offers a commercial license with support and SLA.

Ties to the Mozilla Ecosystem

Thunderbolt complements the previously announced cq platform — a sort of Stack Overflow for AI agents. In the future, agents running through Thunderbolt will be able to exchange knowledge via cq, using locally accumulated experience to improve shared solutions without transmitting confidential data.

Key points:

• Thunderbolt is not SaaS, but a client application with on-prem deployment support.
• Full compatibility with MCP and ACP makes it a neutral hub for heterogeneous AI environments.
• The open MPL 2.0 license allows modification and integration into commercial products.
• Support for mobile and desktop platforms ensures universal access.
• Integration with cq opens the door to decentralized agent learning.

— Editorial Team