Mozilla has launched Thunderbolt, a new AI client designed to help enterprises run their own self-hosted artificial intelligence infrastructure without relying on third-party cloud services. The tool, announced on April 16, 2026, positions itself as a “sovereign AI client” built on top of deepset’s Haystack framework, an open-source platform for orchestrating large language model applications. By integrating with Haystack, Thunderbolt allows users to connect to various AI models and APIs while maintaining control over their data through local storage and optional end-to-end encryption.
The launch reflects Mozilla’s broader strategy to challenge dominant players in the enterprise AI market, including OpenAI and Microsoft, by offering a privacy-focused alternative. Unlike proprietary AI platforms that process user data on external servers, Thunderbolt enables organizations to run models locally or within their own infrastructure, reducing the risk of data exposure. The client supports OpenAI-compatible APIs, allowing compatibility with models such as Claude, Codex, DeepSeek, and OpenClaw, and can integrate with locally stored enterprise data using open protocols. An offline SQLite database serves as a local “source of truth” for the AI to reference during operations.
Thunderbolt is available as native applications for Windows, macOS, Linux, iOS, Android, and the web, with source code accessible via a GitHub repository for those who wish to build it from React components. Mozilla emphasizes that the client supports common AI employ cases including chat, search, research, automation, and cross-device workflows. Optional device-level access controls further enhance security for businesses handling sensitive information.
The underlying Haystack framework, maintained by deepset, provides the modular architecture that enables Thunderbolt’s flexibility. Haystack allows developers to design custom AI pipelines with explicit control over retrieval, routing, memory, and generation—key components in building retrieval-augmented generation (RAG) systems, semantic search tools, and autonomous agents. It supports integration with a wide range of models and infrastructure providers, including Azure OpenAI Service, AWS Bedrock, Hugging Face, and local models, making it vendor-agnostic and adaptable to diverse enterprise needs.
For organizations using Azure OpenAI Service, Haystack offers specific components such as AzureOpenAIGenerator and AzureOpenAITextEmbedder, which require an Azure OpenAI API key and Azure Active Directory token for authentication. These components can be initialized in a pipeline after a PromptBuilder, enabling seamless use of GPT-4, GPT-4 Turbo with Vision, GPT-3.5-Turbo, and embedding models deployed through Azure. Documentation from deepset outlines how to install Haystack via pip and configure environment variables for secure access to Azure endpoints.
Mozilla’s entry into the enterprise AI space with Thunderbolt underscores growing demand for tools that prioritize data sovereignty, and transparency. As businesses turn into more cautious about sharing proprietary information with external AI providers, self-hosted solutions like Thunderbolt offer a way to leverage advanced language models while maintaining compliance with internal data governance policies. The client’s reliance on open-source technology also aligns with Mozilla’s long-standing commitment to open standards and user empowerment.
Industry analysts note that while Thunderbolt does not introduce a new AI model, its value lies in simplifying the deployment and management of existing open-source AI stacks for non-technical teams. By providing a unified front-end interface to Haystack-based pipelines, it lowers the barrier to entry for companies seeking to avoid vendor lock-in. The availability of pre-built native apps alongside customizable source code ensures accessibility across different levels of technical expertise.
Looking ahead, Mozilla plans to continue refining Thunderbolt based on user feedback and evolving enterprise requirements. Future updates may include expanded model support, enhanced analytics dashboards, and deeper integration with identity and access management systems. As of now, the client is available for direct download from Mozilla’s official distribution channels, with detailed setup guides hosted on the project’s GitHub page.
For enterprises evaluating AI strategies that balance innovation with control, Thunderbolt represents a notable option in the growing ecosystem of self-hosted AI tools. Its foundation in proven open-source technology like Haystack, combined with Mozilla’s reputation for privacy-centric software, offers a compelling case for organizations seeking to adopt AI without compromising on security or autonomy.
To learn more about Thunderbolt’s features, supported platforms, and deployment instructions, visit the official GitHub repository for the project. Users interested in Haystack’s capabilities can explore deepset’s documentation for detailed guides on building AI pipelines with Azure OpenAI, local models, and other supported backends.
Share your thoughts on Mozilla’s Thunderbolt and the future of enterprise AI in the comments below. If you found this overview helpful, consider sharing it with colleagues exploring secure, self-hosted alternatives to commercial AI platforms.