Datalayer

An AI agent shouldn't be twelve YAML keys you hold in your head. Our visual editor draws it: the agent in the middle, its model, MCP servers, rules, skills around it. Canvas, spec and Python are one definition. Try it, no sign-in → datalayer.ai #AIAgents #MCP

📐 Agent specs are getting nicer: Oracle's Open Agent Spec, Pydantic AI, runtime-enforced AgentSpec. Datalayer Agentspecs now makes protocols specs too: MCP, A2A, PAP, PACT… declared per agent, checked, compiled to Python + TypeScript 🧩 github.com/datalayer/agentspecs about:blank#blocked #AIAgents

🎉 Personal Agent Protocol 0.11.0: the first complete Python 🐍 + TypeScript 🟦 SDKs, both roles, an example for every feature 🧭 personal-agent-protocol.datalayer.tech/examples/ Now integrating it in Datalayer 🚀 personal-agent-protocol.datalayer.tech/blog/persona...

🎉 Datalayer's Poppy SDKs (Python 🐍 + TypeScript 🟦) are listed on poppy.md! 📚 Must-read: Poppy vs MCP, A2A, UCP, ACP, TAP, PACT. Poppy answers the consent question 🔐 Kudos Fabricio Telles 👏 poppy.md/compare

🎉 Our Personal Agent Protocol SDKs (Python 🐍 + TypeScript 🟦) are on awesome-poppy! Draft 0.1 today; sign-in, conversations and Operations next 🚀 Thanks Javier Chulvi Bernad for the list 🙌 github.com/JaviChulvi/a...

🪐 Jupyter MCP Server 2.2.3 is out—powered again by community contributions. Thanks @kta1kri (safer HTTP), @sachhg (credential isolation), @nickilicious (smaller image responses) & @pma1999 (tool routing)! 🤖 github.com/datalayer/ju...

Follow the datalayer-ai feed bsky.app/profile/did:... to learn more about data analysis

✨ An agent that answers with an interface, not a wall of text. 📅 A form that sends your input back to the agent; 📊 Jupyter output as an A2UI surface, buttons that send your click to the code. 🎬 90 s, 🔓 open source → github.com/datalayer/agent-runtimesabout:blank#blocked #A2UI #Jupyter

🛑 𝙎𝙩𝙤𝙥 𝙥𝙖𝙨𝙨𝙞𝙣𝙜 𝙅𝙪𝙥𝙮𝙩𝙚𝙧 𝙣𝙤𝙩𝙚𝙗𝙤𝙤𝙠𝙨 𝙖𝙧𝙤𝙪𝙣𝙙. Handoffs lose state, files, outputs and context. 🪐 Datalayer gives teams and AI agents one live Jupyter notebook. Same cells. Same state. Same results. datalayer.ai #Jupyter #Collaboration

🚀 𝗝𝘂𝗽𝘆𝘁𝗲𝗿 𝗳𝗼𝗿 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁𝘀 Our new Datalayer homepage is live. Give your agent persistent notebooks, scalable compute & connected data—or try an Agent built for Jupyter directly. The work keeps running when you disconnect. 👉 datalayer.aiabout:blank#blocked

🚀 Jupyter MCP Server 2 is here. Now on MCP Python SDK 2. No breaking change for users. What it unlocks: ⚡ stateless scale 🔐 stronger auth 🔄 richer agent flows 🧩 async + long-running tasks 𝗝𝘂𝗽𝘆𝘁𝗲𝗿 + 𝗠𝗖𝗣 𝗶𝘀 𝗷𝘂𝘀𝘁 𝗴𝗲𝘁𝘁𝗶𝗻𝗴 𝘀𝘁𝗮𝗿𝘁𝗲𝗱.

🚀 𝗧𝗵𝗲 𝗵𝗼𝘀𝘁𝗲𝗱 𝗝𝘂𝗽𝘆𝘁𝗲𝗿 𝗠𝗖𝗣 𝗦𝗲𝗿𝘃𝗲𝗿 𝗶𝘀 𝗹𝗶𝘃𝗲. 🔌 One MCP endpoint gives your AI agent: 📓 Persistent notebooks ⏳ Durable execution ⚡ CPU/GPU compute 🗄️ Direct data access 🤝 Human + agent collaboration No server to operate. 👉 datalayer.ai/docs/mcp

⚡ 𝗝𝘂𝗽𝘆𝘁𝗲𝗿 𝗠𝗖𝗣 𝗦𝗲𝗿𝘃𝗲𝗿 — 𝗛𝗼𝘁 𝗙𝗶𝘅 `code-sandboxes` renamed: 🔄 `jupyter` → `jupyter-server` Use: ✅ `jupyter-mcp-server >=1.5.0` → `code-sandboxes >=1.1.1` ⏪ Older server → `code-sandboxes <=1.0.9` Otherwise: 💥 `Unknown sandbox variant: jupyter` jupyter-mcp-server.datalayer.tech/releases/#ho...

▶️ Run the cell. 🌐 Close the browser. 💻 Close your laptop. Your computation keeps running 🚀 Datalayer for JupyterLab now supports ⏳ long-running cells + 📡 streaming outputs. Try it: pip install "datalayer>=1.7.3" 📚 datalayer.ai/docs/jupyter...

☁️ Remote #GPUs. 💻 Apple Silicon locally. 🤖 Coding agent on your laptop. With 𝗝𝘂𝗽𝘆𝘁𝗲𝗿 𝗠𝗖𝗣 𝗦𝗲𝗿𝘃𝗲𝗿, the agent can drive the remote kernel itself ⚡ “I see the output immediately.” — Ryosuke Tokuami 🚀 datalayer.ai/testimonials

Good time to push the #Jupyter community boundaries

📚 Better docs for Jupyter MCP Server! The MCP tools reference is now automatically generated from source with @sourcey. 👉 jupyter-mcp-server.datalayer.tech/mcp/ Big thanks to @antheducation for the contribution! 🙏 #Jupyter #MCP #OpenSource

Enterprise AI needs more than token metrics. It needs traceability. Know what your agents executed, which tools they called, what data they used, and how they reached a result. That's how AI becomes trustworthy. datalayer.ai/guide #AI #AIAgents #Jupyter

🥈 We just open-sourced our **AI for Earthquake Response** solution. 2nd place at the ESA Φ-lab challenge. 📖 Blog: datalayer.blog/2025/09/12/a... 💻 Code: github.com/datalayer-ch... Next: reproduce it on Datalayer. #AI #OpenSource

**The cheapest token is the one you never send.** Most AI agents keep sending the same context back to the LLM. CodeMode keeps execution inside a **Code Sandbox**, sending only what's needed for the next decision. Fewer tokens. Lower cost. Faster agents. datalayer.ai/guide #AI #AIAgents #LLM

🚀 Jupyter MCP Server v1.2.0 is out. ⚠️ Breaking change: **Runtime** → **Code Sandbox**. This better reflects support for multiple execution backends. It's **not** the MCP v2 migration—that comes later. Migration guide: jupyter-mcp-server.datalayer.tech/releases/#mi... #MCP #Jupyter #OpenSource

AI agents need AI runtimes. Without a runtime, agents repeatedly rebuild context and redo work. Persistent execution makes agents faster, cheaper, and ready for production. datalayer.ai/guide #AI #AIAgents #Jupyter

🎨 One theme doesn't fit everyone. Datalayer now has 5 themes: 🟢 Datalayer (classic) 🌌 Spatial (cosmic) 🌹 Lovely (warm) 💚 Matrix (terminal) 🌍 Earth (nature) Plus Light, Dark & Auto modes. 🎥 www.loom.com/share/4977d2... Which theme would you pick?

🤝 Jupyter MCP Server keeps reaching new users! A recent request is to publish the project on **conda-forge**. If you have conda-forge experience, we'd love your help: github.com/datalayer/ju... New contributors? Start here: github.com/datalayer/ju... #Jupyter #OpenSource #Python #MCP

🪐 Jupyter MCP Server v1.1.3 is out! More reliability, better compatibility, and a smoother experience for AI agents working with Jupyter Notebooks. Release notes: github.com/datalayer/ju... #Jupyter #MCP #AI

A small but important milestone for @datalayer.ai team: Jupyter MCP Server + Code Sandboxes now moves toward a multi-sandbox provider model. Keep one MCP while giving teams freedom to choose the runtime backend that fits each workload. #Jupyter #AI #MCP #Freedom #LockOut

🖥️ Datalayer Desktop 0.0.3 is out! Edit **hosted Jupyter Notebooks** from a native desktop app. No local Python setup—just connect to your notebooks and code. github.com/datalayer/de...

🪐 Jupyter MCP Server v1.0.7 is available. Highlights: • `clear_cell_output` • Better live notebook sync via YDoc • Safer `delete_cell` • Easier installation Thanks to our new contributors @AmirF194 and @sachhg! 🎉 github.com/datalayer/ju...

🧠 Context engineering > prompt engineering. We compare CodeMode vs no CodeMode with reproducible GitHub Actions workflows. Less context → fewer tokens → lower cost. 🎥 Demo: www.loom.com/share/1edbbe... Report deep dive next.

📚 New: Summarize 10 PDFs. Read up to 10 PDFs, extract key ideas, compare themes, and generate a concise summary. 🎥 Demo: www.loom.com/share/45b5ed... 🚀 datalayer.ai/gallery/agen...

📊 New in the Datalayer Agent Gallery: Replace Excel Pivot Work. Ask for the tables, groupings, filters, and charts you'd normally build by hand in Excel—and let the agent generate and export the results by executing Python. #AI #Agents #Excel #Python #Jupyter www.loom.com/share/184f13...

📊 First agent in the Datalayer Agent Gallery! We prepared the classic Titanic dataset as an Excel workbook and let the agent analyze it: answering questions, generating chart by executing Python. datalayer.ai/gallery/agen... #AI #Agents www.loom.com/share/60c6c4...

🚀 Jupyter MCP Server v1.0.3 is out! 🔒 Hardened management API: • localhost-only management routes • Reject non-local browser origins • MCP_TOKEN for state-changing routes ⏱️ execute_code now supports up to 1h timeouts. github.com/datalayer/ju... #MCP #Jupyter #AIAgents

🔒 Your AI platform shouldn't own: ✓ Workflows ✓ Tools ✓ Integrations ✓ Business logic You should. #ManagedAgents #OpenSource #AI

🚀 Managed AI Agents are the future. 🔓 Vendor lock-in doesn't have to be. More this week. #ManagedAgents #AI #OpenSource

The #Jupyter Notebooks MCP Server by @datalayer.ai is a MCP server implementation designed to act as a bridge between Large Language Models, AI assistants and a user's live Jupyter environment.What does it enable? On IProgrammer: dub.sh/mNFIOeI

Big news: @pydantic.dev is donating the FastA2A repo to Datalayer. We'll take over maintenance and evolution of this MIT-licensed A2A framework, building on recent streaming collaboration. Focus: A2UI, Jupyter sandboxes, and agent-to-agent infra. 🚀 datalayer.ai/blog/2026/05...

☁️ Clouder is now open source! Create, manage & share Kubernetes clusters across clouds with a simple CLI & Python package. Includes Helm, autoscaling, backups, and even alpha features like container checkpoint (CRIU). Supports AWS, Azure & OVHcloud. 👉 clouder.sh

This example highlights the power of Agent Runtimes: a single agent can support multiple output formats seamlessly. From plain text and Markdown to structured JSON and UI-ready payloads, responses can adapt dynamically. ➡️ Build once, deploy everywhere. #AI #Agents #DeveloperTools

🛡️ AgentToolApprovalsExample: human-in-the-loop, in 4 lines. id: demo-tool-approvals tools: - runtime-echo:0.0.1 # auto - runtime-sensitive-echo:0.0.1 # manual approval One agentspec, two tools — runtime-echo ships, runtime-sensitive-echo waits for your 👍. #AIAgents #Datalayer

Should your AI agent need approval before running a skill—even for something as simple as telling a joke? What do you think? Personally, I believe it should be configurable 🙂