One of the coolest things about Mastra sandboxes is the ability to FUSE-mount filesystems like S3 Mounts can be read-only for seeding files Or you can give an ephemeral sandbox persistent storage. Even if the sandbox is destroyed and recreated, it can reconnect to the same filesystem and access the same files!

We completely revamped 𝚌𝚛𝚎𝚊𝚝𝚎 𝚖𝚊𝚜𝚝𝚛𝚊 and replaced the humble weather agent with a powerful harness template! It can: - Code in a sandbox - Create a task list - Search and fetch the web - Remember anything - Run for days Quick video showing you how it works:

In July, we shipped 17 features: - real-time notif. - browser recording - custom signal providers - schedules - 1st-class skills - file-based agents - gates + verdicts - tool hooks - goals - memory extractors - agent factory - Trace Intelligence - Envs - Regions - rubric scorers

We've all seen the feedback buttons in AI apps. I wanted to know what actually happens after you click one. Built it with @mastra. Covers the feedback api, the scorers grading every response, and what happens when they disagree.

@mastra thank you so much for having us! Had so much fun talking about self-improving apps and listening to really amazing speakers 😃 If anyone wants to chat about the architecture I spoke about, come say hi or shoot us a message @DefangLabs

Huge TypeScript AI day today! Live from London 09.30-16.30: 9:40 - 9:55 AM: Opening Keynote → Sam Bhagwat, Shane Thomas, Abhi Aiyer, Co-Founders @ Mastra 10:00 - 10:15 AM: Evals Are the New TDD → RL Nabors, Developer Experience @ Arize 10:20 - 10:35 AM: Deploying AI Agents Where Your Customers Actually Want Them → Raphaël Titsworth-Morin, Senior Developer & Evangelist @ Defang 10:40 - 10:55 AM: What I Learned Shipping FGA to Production → Tiffany Lozenski, Founding Product Engineer @ Palette 11:30 - 11:50 AM: The CLI Is Dead, Long Live The CLI → Jan Peer Stöcklmair, Senior Software Engineer @ Sentry 11:55 - 12:15 PM: Agentic UX: How The AG-UI Protocol Is Redefining AI-Driven User Experience → Tyler Slaton, Founding Engineer & Head of Open Source @ CopilotKit 12:20 - 12:35 PM: Building an OpenSearch Research Agent at Scale → Alfonso Graziano, AI Tech Lead @ Nearform 1:45 - 2:00 PM: Why Monetizing Agents Needs Different Infrastructure → John Yeo, CTO @ Autumn → Charlie Lamb, Founding Engineer @ Autumn 2:05 - 2:25 PM: Even TS Agents Need Retrieval → Carly Richmond, Developer Advocate Lead @ Elastic 2:30 - 2:55 PM: From Magic to Harness: Building an AI Agent for a 25-Product SaaS Company → Ilyà Zayats, CTO @ Factorial 3:00 - 3:15 PM: Designing Zero Trust Harnesses → Ohans Emmanuel, Co-Founder @ Coldtea 3:20 - 3:35 PM: Building Cloud Coding Agents on Prisma Compute → Tyler Hogarth, Head of Engineering @ Prisma 3:40 - 3:55 PM: What We Learned Building Our Own Agent Harness → Danilo Delizia, AI Architect @ commercetools 4:00 - 4:20 PM: The Agentic Repository Automation Revolution → Don Syme, Principal Researcher @ GitHub Next 4:25 - 4:30 PM: Closing Remarks → Sam Bhagwat, Shane Thomas, Abhi Aiyer, Co-Founders @ Mastra

Mastra's Agent Builder just got a lot more powerful with our new Composio integration! 💯 Describe the agent you want, and Agent Builder will equip it with the right tools from Composio’s 1,000+ integrations - so you can create more capable agents without defining every tool up front. The legend @YujohnNatt himself put together a quick walkthrough:

You can now define and attach LLM as Judge Scorers in Mastra Studio I can see developers using this to quickly iterate on a scorer before promoting it to code Also handy for non-technical users building with Editor and Agent Builder