Mozilla.ai

What comes after the race to build with LLMs? According to our CEO, John Dickerson, the next challenge isn't the models; it's the infrastructure around them. As LLM applications move into production, routing, governance, reliability and cost control are becoming just as important as model capability. In his latest article, John shares why he believes the next era of AI will be defined by infrastructure and control, not just models. 📖 Read the article: https://bit.ly/4eMW6as #LLM #AIInfrastructure #OpenSource #SoftwareEngineering

𝗛𝗲𝗮𝗱𝗶𝗻𝗴 𝘁𝗼 RAISE Summit 𝟮𝟬𝟮𝟲? Our colleagues Thomas Bille and Caroline Bohu will be there this year and would love to connect with developers, founders and AI teams building the next generation of LLM applications. If you'll be at the event, come and say hello! We'd love to talk about open source AI, LLM infrastructure and the future of AI in production. See you at RAISE Summit 👋 #RAISESummit #AI #LLM #OpenSource #MozillaAI

One theme keeps coming up in conversations with developers and engineering teams: Building with LLMs is no longer the hardest part. Operating them is. It was great to see this discussed during Mozilla.ai's session at Internet Archive Europe, where Davide Eynard and Thomas Bille shared why infrastructure, openness and ownership matter as LLM applications move into production. If you're interested in where AI infrastructure is heading, this is a great recap. Thanks to everyone involved!

Otari is now generally available. As LLM applications move into production, managing multiple providers, routing logic, budgets and governance has become a growing engineering challenge. Otari is an open-source LLM Control Plane built by Mozilla.ai, providing developers with a single layer to manage and control their LLM infrastructure. We're excited to share what we've been building and would love your feedback from the community. 📖 Read the launch article: https://lnkd.in/eUzSfBs9 #LLM #OpenSource #AIEngineering #DeveloperTools #MozillaAI

Speech-to-text should be as portable and easy to deploy as today's best local AI tools. Meet transcribe.cpp: a new open-source C/C++ speech-to-text inference library developed through Mozilla.ai's Builders in Residence program. Built on ggml, transcribe.cpp provides: 🎙️ Support for multiple STT model families via GGUF ⚡ GPU acceleration with Metal, Vulkan, and CUDA 💻 Portable, local inference across platforms 🔓 An open foundation for building fast transcription into your own applications transcribe.cpp is designed to solve a common challenge: many excellent speech-to-text models exist, but they're often tied to specific platforms or lack optimal performance. transcribe.cpp provides a unified interface that makes audio transcription easier to build and deploy. 👀 Read the full article: https://lnkd.in/e2agbERF Let us know what you'll build with it! #OpenSource #AI #SpeechToText #MachineLearning #Cpp #MozillaAI

Product management is often less about building features and more about managing information. Feedback lives in Slack. Decisions are buried in Notion. Pull requests are on GitHub. Context is scattered across dozens of tools. In our latest blog post, we share how we use Octonous inside Mozilla.ai to reduce that overhead, not by replacing our existing tools, but by connecting them. Some of the workflows we rely on every day include: ✅ Turning Slack feedback into GitHub issues with a simple emoji reaction. ✅ Receiving product-focused summaries of merged pull requests. ✅ Searching across connected tools from a single place instead of hunting through tabs. The goal is to spend less time on repetitive work and more time building products. 👀 Read the full article and let us know how you streamline your product workflows: https://lnkd.in/da3RvW2e #MozillaAI #Octonous #AI #Automation #ProductManagement

Heading to ACM FAccT 2026? Don't miss this hands-on tutorial today! Two of our brilliant minds at Mozilla.ai, Roya Pakzad and Daniel Nissani, are leading an essential session on evaluating trust, safety, and guardrails in the age of advanced AI. 🗓️ Today, June 25 | 5:00 PM - 6:00 PM EDT (Montreal Time) 🔍 Topic: Contextual Evaluation of LLM Guardrails Across Languages and Agentic Systems As LLMs become deeply integrated into our society, evaluating the guardrails that protect them is just as crucial as evaluating the models themselves. Roya and Daniel will be diving deep into some of the most pressing questions in AI safety today: 💡 Why guardrails must be contextual: Safety taxonomies and policies need to naturally fit the specific language, domain, and deployment setting. 📂 The open-source opportunity: How the rise of open-source and LLM-based guardrails opens the door for rigorous, independent evaluation. 🤖 Agentic guardrails: Can guardrails themselves leverage web search, API calls, and retrieval tool-use to ensure fact-checking and trustworthiness? 💻 This is a hands-on session! If you are attending FAccT, be sure to bring your internet-connected device to participate. Can’t make it in person? Explore the research, web application, and methodologies behind their work through the links below: 📝 Mozilla.ai Blog: https://lnkd.in/eccKGvtv 📅 ACM FAccT Program Details: https://lnkd.in/eHWMiMcg #FAccT2026 #AISafety #LLMGuardrails #ResponsibleAI #OpenSourceAI #MozillaAI

Big news from the Mozilla.ai team! Tomorrow, two of our engineers, Davide Eynard (ML Engineer) and Toto Thomas Bille (Software Engineer), are taking the stage at the Internet Archive Europe event right here in Amsterdam. They’re going to be talking about something we’re incredibly passionate about: "Mozilla.ai - Own your AI agent: running open source agents on your terms." If you've been wanting to move away from rigid, proprietary AI setups and figure out how to actually deploy and control open-source agents on your own terms (and with total privacy), you won't want to miss this. The logistics: 📍 Where: Oudeschans 16, Amsterdam (or right from your couch) ⏰ When: June 25 at 2:00 PM Amsterdam time ⚠️ Quick heads-up: In-person spots are officially sold out, but you can still hop on the Luma waitlist. That said, you don’t have to miss a thing—the whole session is being streamed live. Pop by and say hi in the Zoom chat! 👉 Grab a waitlist spot: https://luma.com/rfujxz40 👉 Tune in live tomorrow: https://lnkd.in/eyPD93JU See you there! 🦊 #Mozilla #OpenSource #AI #InternetArchive #AmsterdamTech 

Image classification is now available in encoderfile 📸 encoderfile makes it easy to deploy encoder models as single executables. No Python runtime, no dependencies. Until now it only worked with text: embeddings, classification, sentiment analysis. Now it handles images too! This release introduces: ✅ Image classification via CLI, HTTP/S, and gRPC ✅ Flexible preprocessing with Lua scripts (resize, normalize, channel handling) ✅ A foundation for upcoming object detection and image segmentation 👀 Read the full blog post: https://lnkd.in/gCTMXsw8 What features you'd like to see next? Let us know! #AI #MachineLearning #ComputerVision #OpenSource #MLOps #MozillaAI

Hasan Gözlügöl’s latest thoughts on Octonous capture how AI can empower teams to work more securely and efficiently. It’s exciting to see how innovation at Mozilla.ai is driving meaningful change across our products. Insightful perspectives like these encourage us to keep pushing the boundaries of what’s possible.