Open-source AI for developers. Building tools that foster transparency, accessibility, and real-world impact.
Product teams often spend hours each week answering repeat roadmap questions.
We’re exploring early concepts for the Agent Platform, including a Roadmap Agent that turns roadmap details into clear, easy-to-read answers.
Learn more: https://link.mozilla.ai/roadmap-agent
📢 We’re rolling out the alpha of the any-llm Managed Platform!
It offers zero-knowledge API-key storage, unified cost tracking, budgeting controls, and usage analytics that never expose prompts or responses.
It also integrates directly with any-llm and any-llm-gateway.
Full blog post: https://link.mozilla.ai/any-llm-managed-platform
Do you choose an LLM based on cost, accuracy, or response time?
Or do you switch between models depending on the query?
If you want a clearer way to compare them, here is our blog post: https://blog.mozilla.ai/the-challenge-of-choosing-the-right-llm/
The recording of Davide Eynard’s (@mala) “Build Your Own Timeline Algorithm” talk from SFSCON is now live.
In the session, Davide walks through how to build a local, customizable timeline algorithm using https://Mastodon.py, llamafile, and marimo.
Watch the full video: https://www.sfscon.it/talks/build-your-own-timeline-algorithm/
We just wrapped our sessions at DataFest Tbilisi (Nov 27–29)!
• Raz Besaleli on systems engineering lessons for more robust AI
• Davide Eynard (@mala) on running your own open-source AI agents
• Irina V. on treating AI software with the same rigor as any other system
Thanks for being part of it — see you next year!
We’re opening up some early concepts from the Agent Platform.
The Agent Showroom highlights a few prototype agents, including call prep, post-mortem drafting, and roadmap support.
Learn more: https://link.mozilla.ai/agent-platform
Why developers use any-llm:
• One interface for every LLM provider
• Cleaner integrations with fewer rewrites
• Consistent behavior across models
Install it here: https://link.mozilla.ai/any-llm-repo
DataFest Tbilisi starts tomorrow 🎊
From Nov 27-29, Raz Besaleli, Davide Eynard (@mala), and Irina Vidal will be presenting on AI resilience, software reliability, and practical ways to run open source AI agents with full control.
More info: https://www.datafest.ge/
Encoderfile v0.1.0 introduces a single-binary deployment format for encoder transformers.
No containers and no runtime dependencies, just deterministic, self-contained binaries built with ONNX and Rust.
Learn more: https://link.mozilla.ai/encoderfile-v010
The lethal trifecta creates new failure points in agent systems.
Most guardrails miss them.
Read our full analysis: https://link.mozilla.ai/open-source-guardrails
If your team works with MCP tools, here’s something new that may simplify your workflow.
mcpd-proxy gives developers a single access point for all MCP servers in VS Code, Cursor, and other IDEs. No more manual setup or mismatched configs.
We’ve been exploring a cleaner way to work with MCP tools inside IDEs like VS Code and Cursor.
A small update is coming tomorrow that we think teams will find useful.
Big cloud or local models. Same interface.
any-llm (v1.0) keeps your workflow stable.
👉 Try it today: https://github.com/mozilla-ai/any-llm
Local LLMs are back for good reason:
✅ Control
✅ Privacy
✅ Simplicity
llamafile makes it easy. One file, no setup.
Start using llamafile: https://github.com/mozilla-ai/llamafile
Introducing any-llm-gateway 🏗️
A FastAPI-based open-source gateway to manage and monitor LLM usage across providers.
Handle budgets, track costs, manage API keys, and deploy with confidence all in one place.
Read the full post: https://blog.mozilla.ai/control-llm-spend-and-access-with-any-llm-gateway/
#MozFest 2025 has wrapped up after three days of creative discussions and collaboration around a better internet for everyone and open-source AI.
Grateful to everyone who joined us and to the organizers for an incredible experience.
See you next year!
#MozFest 2025 🎉
Creators and communities are gathering to explore how technology can better serve people.
@peteski22, Javier, and Mario from Mozilla.ai are attending, and our CEO John is co-hosting “Context is All You Need: The Collective Land Grab Reshaping the Internet.”
We put open-source guardrails to the test for agent safety.
Setup:
• Indirect prompt injection: BIPIA email/table + benign WildGuardMix
• Function-call malfunctions: HammerBench
Results:
• PIGuard was effective on injection detection
• Custom judges struggled on function-call correctness
Methods, data, and code: https://blog.mozilla.ai/can-open-source-guardrails-really-protect-ai-agents/
Your feed should serve your curiosity, not engagement goals.
Join Davide Eynard (@mala) from Mozilla.ai at SFSCon to explore how to build a personal, local timeline algorithm that runs entirely on your computer using Mastodon.py, llamafile, and marimo.
🕔 November 7 at 17:00 (Seminar 4)
🔗 https://www.sfscon.it/talks/build-your-own-timeline-algorithm/
🚀 Any-LLM v1.0 is now live!
One API for every model whether cloud-based or local. Run OpenAI, Claude, Mistral, or llama.cpp.
Version 1.0 brings better performance, stability, and a cleaner developer experience.
Read the full post: https://blog.mozilla.ai/run-any-llm-with-a-single-api-introducing-any-llm-v1-0/?utm_source=mastodon
