Yesterday we announced the open preview for LlamaAgents So, me and my colleague Adrian wrote this intro blog to help you get started. Learn about all the document agent templates available to you via llamactl and:

LlamaExtract now has a PER_TABLE_ROW extraction target. Instead of extracting once per doc or per page, you can now apply your schema to each row in a table or each item in a bulleted list—getting back an array of structured JSON objects.

Last week, I spent some time on the DeepSeek OCR paper with @LoganMarkewich and it was genuinely interesting. It's definitely not just OCR - it's using vision as compression. 10x token reduction while maintaining accuracy.

Build Text-to-SQL systems with open source Arctic models and agentic workflows 🚀 Our own @tuana.dev is speaking at Snowflake Build to show you how to create powerful data analysis agents that can handle complex, multi-part questions:

An interesting overlap exists between memory and context. And a new(ish) memory block that was added as an integration to @llama_index is (imo) - a great demonstration of that 👉 the Artifact Memory Block

This past Monday I had the pleasure of joining a panel about open-source and AI, hosted by Merantix at the London AI Hub.

We're opening up early access to LlamaAgents - which allows you to go from local LlamaIndex agent workflows to deployed agents in a matter of minutes.

Wrote about something I've been thinking about lately: how AI coding agents are changing how we build AI agents themselves At LlamaIndex, we've been experimenting with using coding agents like Cursor and Claude Code to speed up build agentic workflows themselves + their UIs

We're making 𝐯𝐢𝐛𝐞-𝐥𝐥𝐚𝐦𝐚 with and for LlamaIndex a lot easier.. Last week we released vibe-llama, a tool developed by @cle-does-things.bsky.social

This is not GraphRAG, it comes before.. Last week, me and Tomaz (from neo4j) published another demo together. The idea: you have a bunch of unstructured documents (think legal, PDFs etc) and you want to construct a knowledge graph out of them.

OpenAI just dropped new open source models 🔥 My colleague Logan spent his morning testing gpt-oss-20b - which runs on consumer hardware, is Apache 2.0 licensed, and the reasoning quality is solid. Also, full chain-of-thought access so you can see exactly how it's thinking.

Ok this is prettyyy cool, another ‪@llamaindex.bsky.social‬ workflows implementation from @cle-does-things.bsky.social 👇

Last week @cle-does-things.bsky.social and I created @llamaindex.bsky.social nodes for @n8n.io‬ Here's what we got: · LlamaExtract: use extraction agents within n8n workflows · LlamaParse: parse complex documents · LlamaCloud: use your existing LlamaCloud indexes as knowledge bases within n8n

A new walkthrough for a research agent using Llamaindex and Google Gemini, fresh out the oven. Given a topic: 🌎 Use Gemini 2.5 pro with its server side google search tool 📝 Create an agent that takes notes as it gets results from its websearch

Watch as Claude acts as a document agent using LlamaCloud MCP tools to: - Query any index with custom parsing. - Use LlamaExtract agents to pull structured data based on LlamaCloud's schema. The demo features extraction agents for invoices and technical resumes. youtu.be/IL3CEONiDF4

With all of this "context engineering" chatter, I've been thinking about how much of LLM performance hinges not just on model architecture, but on how you feed it context a lot lately. So, co-authored a piece on Context Engineering with Logan: Here it is: 🔗 www.llamaindex.ai/blog/contex...

I like this new terminology around context engineering that's picking up. It's a good way to think about how to build actually effective agentic applications.

Today we're announcing the first stable release of @llama_index workflows. A lightweight framework for building complex, multi-step agentic AI applications in Python and Typescript www.llamaindex.ai/blog/announ...

Since MCP there are questions about whether it removes the need for you to build your own RAG workflows with vector search. And my answer to that is:

Last week, we concluded the @gradio-hf.bsky.social‬ MCP hackathon with @hf.co‬. The project that one the @llamaindex.bsky.social prize was the "Nasa Space Explorer" 🔭🪐 3 servers that provide live data on: ☄️ Asteroids 🤖 the Mars Rover 🌌 Astronomy Here's the space: huggingface.co/spaces/Agen...

Yesterday Zoom announced Zoom RTMS at the developer summit, which gives you real-time access to audio, video, and transcript data from Zoom Meetings.

Tomorrow, I'm joining Laura for a discussion on memory. Note: not all applications that are 'agentic' necessarily _need_ memory. And what type of memory you need also depends on the use case. Join us at 10AM PT/7PM CET: lu.ma/t27lryii

Order Completion Agent with Artifact Editor by @tuana.dev: Fill out structured forms as the user talks! Checking the completeness along the way: ✅ ArtifactEditorToolSpec to define structured artifacts ✅ ArtifactMemoryBlock to track progress docs.llamaindex.ai/en/stable/e...

🍕 Since the first @llamaindex.bsky.social‬ office hours with Logan a specific use-case kept popping up: incremental "form filling". So, here's a pizzeria order taking assistant using just that with the new Artifact Memory Block and Artifact Editor Tool: colab.research.google.com/drive/13ORO...

We’ve started to host more regular office hours now. Come along to ask anything LlamaIndex 👇

I’m hosting a discussion session about the biggest blockers to productionizing AI agents at Snowflake Dev Day today. Come by to the Group By() space at 2PM to have a chat 🩵

A new oss image generation agent by mu colleague Clelia, that doesn't _just_ generate images. Structured outputs with Google Gemini to check the generated image and evaluate whether it's ready for the user or not. The repo: github.com/run-llama/i... Her code-along: youtu.be/RCASiu5oj6c...

@tuana.dev's new code walkthrough shows how to implement citations and reasoning in LlamaExtract. Learn to define schemas, add citation capabilities, and understand agent reasoning for extractions. Watch here: https://youtu.be/01kM7tXRHi4

LlamaExtract (by @llamaindex.bsky.social) allows you to turn on citations and reasoning to check: 1️⃣ The underlying models reasoning for the assignment and 2️⃣ The citations from the source document 📚 www.llamaindex.ai/blog/get-ci... 🧑‍🍳 docs.cloud.llamaindex.ai/llamaextrac...

Numerous ways to build agents, numerous workflow patterns and applications. On April 29th we will be hosting a live session where we dive into @weaviate.bsky.social agents, how you can use them, and all of our future agentic plans! Join us 👇 www.linkedin.com/events/73157...

Build a personal shopping assistant in less than 5 minutes with the new @weaviate.bsky.social Query Agent! 👩‍🍳 To get started, try out the recipe! You’ll end up building an e-commerce assistant agent: colab.research.google.com/github/weavi...

In 2 weeks I will be attempting to “run” the Cambridge Half Marathon with my mum and aunt again and am raising funds for cancer research while doing so 🩵👇 PS: No I don’t have a pace aim.. I will crawl if necessary. fundraise.cancerresearchuk.org/page/tuanas-...

There are valid concerns when it comes to how we use agents today and how we should prepare to use them in future. 1️⃣ AI agents as ‘research assistants’, 2️⃣ Agentic workflows, some involving human-in-the-loop systems. 3️⃣ Safety nets, such as ‘undo’.

Agents can seem a bit like wizardry, so I've been working on an intro 💚 🧙 The way I like to explain it is by comparing most LLMs in agentic AI applications today to a wizard with a spell book (tools). It's been granted instructions for bunch of spells 🪄 Read on here: weaviate.io/blog/ai-agents

🙌 Join me this Thursday in London during a founder event where I’ll showcase @anthropic.com + @weaviate.bsky.social and where you can see other speakers from Raycast and Lovable! 👉 lu.ma/anthropicLON...

I built a 'game recommender service' with the new @deepseek.bsky.social model using Ollama 🦙 and @weaviate.bsky.social 😈 Here's the recipe and the fancy graphic I created for it 🧑‍🍳: github.com/weaviate/rec...

Stepping out of my usual content here and jumping in to Greek mythology. I’ve gotten really into it in the last few years and just finished Stone Blind by Natalie Haynes. Beautifully written. Would highly recommend this retelling of the story of Medusa.

2 pieces of content on the topic of LLM powered agents that I'd recommend: - The latest Latent Space episode by @swyx.io with @simon.fedi.simonwillison.net.ap.brid.gy - highly recommend! - The GoEX research paper. Late to the party on this one but the contents of it is more and more relevant today

I have a new hobby that involves lots of fire 🔥 Behold my first ring that has a stone setting 💍

Gave my first first talk at @weaviate.bsky.social this past Wednesday 🚀 We created a mini Weaviate assistant agent by defining our own 2 tools: · A Weaviate Docs RAG tool · A Weaviate GitHub Issues RAG tool I used @llamaindex.bsky.social for this demo Here's the repo: github.com/TuanaCelik/w...

I'm preparing a talk about what we mean by 'agentic AI' these days. And I think in the simplest terms I can explain it as: any AI where the burden of making decisions on 'what next' belongs to an LLM that has access to resources to resolve queries. 🧑🏻‍✈️LLM take the wheel.

The best swag I ever got was from the GitHub HQ in SF. Grateful for this little ball of lip balm in the cold Amsterdam weather today.

🇳🇱 If you're in Amsterdam, Wednesday the 27th I'm talking at the Agentic AI Meetup 🚀 My plan is to discuss: 1️⃣ What we mean by 'Agentic AI' 2️⃣ All the different interpretations of term agentic 3️⃣ The different architectures you can adopt 👉 Registrations are open: lu.ma/izoaou7c

Friends, I'm trying to understand the different understandings people have on this (I definitely have...opinions on this) 👇 What constitutes an 'Agentic' application in AI? What makes something 'agentic' in your view?

“Vanilla RAG” and agentic RAG are not mutually exclusive. Vanilla RAG is still a very useful and relatively lightweight method which in itself can be used as tools ⚒️ for agentic applications. I don’t see a one or the other transition happening. I see it as one becoming a resource to the other 🙌

I’m new here 👋