Mozilla’s open-AI roadmap

Mozilla is trying to turn AI into something you own, not something you rent by the month from a handful of cloud companies. In Mozilla’s January 2026 strategy post, Raffi Krikorian wrote that today’s closed systems win because one company bundles the model, hosting, guardrails, monitoring, and billing into a single easy package. (blog.mozilla.org) That sounds abstract until you look at what Mozilla is actually building. Mozilla.ai’s current product page is full of plumbing for developers: any-agent to compare agent frameworks, any-llm to switch model providers, any-guardrail to swap safety layers, mcpd to manage Model Context Protocol servers, and llamafile to run models locally as a single file. (mozilla.ai) The bet is that AI will become a new middle layer between you and the web, the way a browser once was. Mozilla’s January post calls this “Layer 8,” meaning software agents will increasingly filter information, negotiate on your behalf, and decide how you interact with websites and services. (blog.mozilla.org) That is why Firefox keeps showing up in Mozilla’s AI plans even when the company is not pitching a giant frontier model of its own. Mozilla’s Firefox team said in late 2025 that AI is becoming a more widely adopted interface to the web, and argued that an independent browser is one of the few places that can still defend transparency, accountability, and user agency. (blog.mozilla.org) Mozilla is also using the word “sovereign” very deliberately. In its February 13, 2026 post before the India AI Impact Summit in New Delhi, Mozilla said countries face a choice between “renting” AI from a few global corporations or investing in open-source systems that can be inspected, improved locally, and governed in the public interest. (blog.mozilla.org) That language turned into a concrete partnership on March 26, 2026. Mozilla announced its first partnership with a major artificial intelligence research lab, Mila in Quebec, to work on open-source and sovereign AI, starting with private memory architectures for AI agents. (blog.mozilla.org) Private memory sounds technical, but the idea is simple: if an AI assistant remembers your preferences, calendar habits, or work context, Mozilla wants that memory handled in ways that are more inspectable and less dependent on a closed vendor. Mozilla and Mila said the gaps in open AI are no longer just models and compute, but the layers that decide whether systems are private, trustworthy, and usable across many languages and cultures. (blog.mozilla.org) This is also why Mozilla’s roadmap looks more like a hardware store than a finished house. Mozilla.ai says its tools are meant to give developers a “unified foundation” for orchestration, model selection, evaluation, and local execution, which is another way of saying Mozilla wants to make the open stack feel less like loose parts scattered across 20 repos. (mozilla.ai) The company’s own framing is that this is the same role Firefox played in the 2000s. Mozilla’s December 2025 strategy post said Firefox helped break Microsoft Internet Explorer’s 95 percent browser share and pushed open standards into the mainstream; Mozilla now says it wants to do for AI what it once did for the web. (blog.mozilla.org) Mozilla’s 2025–26 annual report makes clear this is not a side project. The report says Mozilla is investing in open-source AI and privacy-preserving technology while keeping its “double bottom line” of public interest plus sustainable business, which is a clue that the company sees developer tools and trusted consumer products as the business model around the mission. (blog.mozilla.org) So the roadmap is less “Mozilla will beat OpenAI with one giant model” and more “Mozilla wants the roads, guardrails, and street signs to stay open.” If that works, Firefox becomes the user-facing layer, Mozilla.ai becomes the tool layer, and “sovereign AI” becomes Mozilla’s pitch to developers, institutions, and governments that do not want their entire AI future leased from someone else. (blog.mozilla.org)