| Website | https://www.mozilla.ai/ |
| Agent Platform | https://www.mozilla.ai/product/agent-platform |
| Choice-First Stack | https://www.mozilla.ai/open-tools/choice-first-stack |
We’re hosting the Octonous workshop in Lisbon 🇵🇹 today.
It’s a hands-on session where you’ll build an AI agent for a real task.
Bring a task and leave with a working AI agent.
No technical skills needed.
A few seats are still available.
Join here: https://link.mozilla.ai/octonous-lisbon
Daniel Nissani (AI Safety Engineer) and Roya Pakzad (Senior Mozilla Fellow) evaluated three guardrails on multilingual humanitarian prompts:
• FlowJudge
• Glider
• AnyLLM (GPT-5-nano)
Each reviewed the same asylum-related scenarios in English and Farsi.
A technical evaluation of multilingual, context-aware AI guardrails, analyzing how English and Farsi responses are scored under identical policies. The findings surface scoring gaps, reasoning issues, and consistency challenges in humanitarian deployments.
https://Mozilla.ai is now a launch partner of Flower Hub.
We published a new project called **fed-phish-guard**, which trains a phishing URL classifier using federated learning.
Each client trains locally and only sends model updates back to the server, keeping browsing data private.
Read the blog post here: https://link.mozilla.ai/flower-hub
Check the project page here: https://flower.ai/apps/mozilla-ai/fed-phish-guard/
LLM providers differ in streaming behavior, error semantics, and supported features.
any-llm-go normalizes those differences behind a single Go interface so applications can work across multiple providers.
Take a look at the repository: https://link.mozilla.ai/any-llm-go-repo
BYOTA combines a few tools to experiment with timeline algorithms.
https://Mastodon.py retrieves posts from Mastodon timelines.
llamafile runs language models locally.
marimo provides an interactive notebook in the browser.
Watch the full talk: https://link.mozilla.ai/sfscon-byota
For decades, engineers produced code at roughly the speed they could understand it.
AI changes that dynamic.
Now a developer can generate far more code than they can reason about line by line. Reliability starts shifting into the surrounding system.
How are teams adapting their engineering practices?
Full post: https://link.mozilla.ai/owning-code-ai
Every model has blind spots.
The Star Chamber sends code review questions to multiple LLM providers at once and aggregates the feedback by agreement level.
Consensus issues come first. Majority next. Individual observations last.
A practical way to review code from multiple perspectives.
The Star Chamber runs code reviews across multiple LLM providers and aggregates their feedback by consensus. Instead of relying on one model’s perspective, developers get a structured view of where models agree, disagree, and raise unique insights.
The real bottleneck for many solo contractors is the paperwork.
Nathan Brake, Senior ML Engineer, is leading Clawbolt: a messaging-first AI assistant built for teams of one.
Estimates, memory, voice memos, photo documentation, reminders.
Check out our latest project: https://link.mozilla.ai/clawbolt-ai
Many timeline algorithms share the same basic steps:
1. Moderation
2. Candidate generation (posts you’re likely to like)
3. Ranking and re-ranking (changing the order)
It’s a simple structure, but it shapes what you see.
Watch the full session: https://link.mozilla.ai/sfscon-byota
https://Mozilla.ai is headed to SCALE 23x in Pasadena.
SCALE takes place March 5 to 8, 2026 and is North America’s largest community run open source conference.
Anushri, Raz, Nathan, and John will be there.
Visit us at booth #126 and say hello. Join us in Pasadena: https://link.mozilla.ai/scale23x-2026
