Release notes: Docs (serving): Docs (tools):

New any-agent release (v0.24.0) is out. ▪️ AgentJudge returns full eval traces.▪️ Serve agents via MCP.▪️ Use agents as tools.▪️ Cost + token tracking.▪️ A2A protocol compatibility. We're making agents more composable + observable. ⬇️ Links in next post

🌐 New blueprint: AI agents that run in your browser via Webassembly & Pyodide. Just open an HTML file and go. They work with OpenAI API compatible models or local models via Ollama/vLLM. Take a look :

🚀 Any Agent 0.22.0 is here with AgentJudge and LlmJudge!. Two new ways to evaluate your agents:. LlmJudge for qualitative assessments.AgentJudge for complex evals with tool access. Build more reliable agents with better evaluation primitives. Docs:

🎯 New Blueprint: Speech-to-text alignment.Tired of Whisper butchering your name or technical terms?. With Michael Hansen from the @openhomefndn , we built a tool to align any Whisper model to YOUR words. Take a look :

🚀 Any Agent v0.21.0 is here!. Now with multi-turn conversations using Google's A2A protocol. Build sophisticated multi-agent systems where agents maintain context and collaborate across exchanges.

🚀 The BYOTA (Bring Your Own Timeline Algorithm) demo from FediForum 2025 is now live!. Watch how users can take control of their social media algorithms for a more transparent, customizable experience. Demo: sessions:

Any Agent 0.19.0 is out! 🚀. Key updates:. Structured Outputs - Custom Pydantic schemas for reliable responses.Multiagent Systems via A2A - Compose agents as tools.Unified OpenTelemetry Tracing - Standardized observability. Release notes: Check it out!.

🎤 @johnpdickerson (our CEO) will speak at the @aiDotEngineer World's Fair in SF this week!. His talk explores why we keep saying "this is the year of evals" - examining what actually drives evaluation adoption vs. what should. 📍 Attending the conference? Let's connect!.

🚀📄 New Community Blueprint: LLM Document Parser. Built with @CSUMB students, Transform docs/PDFs into structured data using OCR + LLMs. ✨ Upload → OCR → LLM parsing → Clean JSON. Perfect for bank statements, invoices, forms. Runs locally. 🔗

Ready to build AI agents? Here’s the first Any-Agent Cookbooks!.Cookbook 1: Build your first AI agent.Cookbook 2: Create an interactive booking assistant with MCP. Explore the cookbooks:.- -

New Blueprint: Test and compare AI agents using a real-world use case: finding the best surf spot!. 🔄 Switch between agent frameworks.⚙️ Test different agents & models combo.📊 Evaluate performance. Demo: Code:

Generative AI is powerful but unpredictable. Small prompt changes can cause inconsistent or biased results. We tested DSPy for prompt optimization, and built a prompt saliency tool to understand LLM behavior. 📖 Read more:

🧠 Choosing the right LLM is hard. Our latest post explores how LLM Routing helps devs intelligently direct queries between models based on accuracy, speed, or cost . Read more on our blog :

🎥 Watch the x @AiEleuther webinar recording! See experts demo the Audio Transcription & Document Conversion Blueprints, including setup, best practices, and tips. Recording: Explore more:

We're launching our first-ever Builders in Residence Program for practical, application-driven AI researchers (PhDs, recent grads, industry). Fully remote (EU, US, CA), ~3-month fellowship starting June 2025. Learn more & Apply here:

🧪 Testing AI ≠ testing software. From hallucinations to prompt injections and data leaks, LLMs introduce new risks. Traditional testing won’t cut it. We unpack the challenges (and tools) in our new blog post:

New Lumigator Updates!.✅ Provided better translation support with Opus-MT models and G-Eval integration. ✅ Expanded evaluator capabilities by integrating DeepEval's Ollama. ✅ Refreshed the Experiments UI with new features and improvements. Try it out:

Are AI apps really tested? Most devs rely on manual checks. So we built AIssert: a lightweight tool to test LLM-powered apps built on @giskard_ai .✅ No Python needed.✅ No manual LLM config .✅ Plug into any CI/CD. Repo: Blog post:

🚀 New! @MozillaAI x @AiEleuther Blueprints for easier & open dataset creation. 1️⃣ Audio Transcription using privacy-focused Whisper models. 2️⃣ Document Conversion to Markdown for building open-text datasets . Available now :