Tweets

Make more expressive videos.
We’ve enhanced the way Veo 3.1 uses reference images to help you create videos that are more expressive, maintain more consistency and give you more creative control.

Outputs are more dynamic and engaging. Characters stay the same when the setting changes. Scenes stay the same if you transform subjects within them. And combining disparate elements (characters, objects, textures and stylized backgrounds) is completed with more finesse. All with shorter prompts.

We closed 2025 at over $330M ARR - driven by our customers’ trust and impact.

@matiii joined @Bloomberg earlier today to talk about the work we do with enterprises, our recent research release, and what’s next.

We’re proud to announce our partnership with @deutschetelekom to bring ElevenLabs' AI voice agents to the customer service of Europe’s largest Telco (via app and phone).

Deutsche Telekom customers will soon experience realistic AI voice agents - available 24/7 and without any waiting time - to augment customer support with a more personal, human-like interaction.

This initiative is part of a shared vision to democratize access to AI, making advanced voice technology simple, accessible, and valuable to everyone.

I made a simple document extraction agent that can extract every single chart from a 73-page ICONIQ report on SaaS startups in 2025, of which 2/3 of the slides have one or more charts 📈📊. We automatically extract each chart into a 2D table.

First off, these slides are super useful on their own: firmographics, revenue multiples, burn multiple, ARR per FTE, magic number etc. All great benchmark data for founders :)

Large scale structured extraction over documents is notoriously hard. If you wanted to automate understanding over these docs, ChatGPT + structured outputs won’t cut it.

LlamaExtract lets you use VLMs to process all types of charts per page, and allows you infer data where labels don’t exist. The results are all joined together at the end for you to export.

Also, this took me ~5 mins to setup. You can create the schema through natural language!

Check out the video below

Sign up to LlamaCloud herellamaindex.ai/signup67

Files are becoming the primary interface for AI agents to manage context, store conversations, and access skills 📁

@jerryjliu0 breaks down how coding agents like Claude Code and @cursor_ai are centralizing around filesystems as core abstractions, moving away from complex tool ecosystems:

📝 Agents store long conversation histories in searchable files to overcome context window limitations
🔍 File-based retrieval with semantic search outperforms traditional RAG patterns for dynamic context traversal
⚡ Skills defined as simple files are replacing complex MCP tools - just copy API specs into markdown files
🛠️ Agents need only ~5-10 core tools (CLI, code interpreter, web fetch) plus filesystem access to be highly capable

The challenges ahead include parsing non-plaintext documents (PDFs, Word, Excel) and scaling file search to massive collections. That's exactly why we built LlamaCloud's Parse, Extract, and Sheets capabilities - to convert any document format into agent-ready context.

Read the full analysllamaindex.ai/blog/files-are…zUc6

Roe AI (@try_roe_ai) is building Rori, a Cursor-like tool for risk and compliance.

Rori acts like a senior analyst. It pulls data from internal tools and public records, connects people, companies, transactions, and behaviors, and produces audit-ready evidence, saving teams millions of hours.

roe-ai.com/blog/a-new-era

What is needed for startups and medium-size tech companies to contribute to open science and open-source AI more?

When thinking about open-source, people usually think about big tech or academia but in my opinion, startups and medium-sized tech companies could be massive contributors and benefit a ton from sharing AI models, datasets, research,... (in terms of visibility, hiring, ability to transition into AI,...). 

We're seeing that a lot in China and looks like we might start to see it in the US too as showed by the fact that the two trending models on @huggingface have been from these types of orgs (@fal and @Lightricks). 

Also interesting that @Airbnb @bchesky hired @Ahmad_Al_Dahle (former Llama lead) maybe to do more of that?

You can now use AWS databases in v0.

Go from prompt to working app with a production database in minutes. No manual set up required.

Learn more: vercel.com/blog/aws-datab…

Constellation (@constspace) predicts satellite network failures before they happen and automatically reroutes traffic in seconds—no human intervention, no data loss.

Congrats on the launch @KamranMajid11, @raaidkabir, @laithaustin, @omeedtehrani!

ycombinator.com/launches/P9f-c…

Very cool