Content History
RT Clelia Bertelli (🦙/acc) New blog post alert!🦙✨ At @llama_index we shipped LlamaAgents Builder to make it the fastest way to build a document agent only with natural language. Yesterday, we published a full walkthrough of how to use LlamaAgents Builder end-to-end with a real-world use case, covering prompt and context engineering best practices, workflow visualization tricks and tips on how to best iterate with the builder over multiple turns. Read the blog: https://www.llamaindex.ai/blog/creating-a-deal-sourcing-agent-with-llamaagents-builder Original tweet: https://x.com/itsclelia/status/2027498003851055531
RT LlamaIndex 🦙 Turn your PDF charts into pandas DataFrames with specialized chart parsing in LlamaParse! This tutorial walks you through extracting structured data from charts and graphs in PDFs, then running data analysis with pandas - no manual data entry required. 📊 Enable specialized chart parsing to convert visual charts into structured table data 🐼 Extract table rows directly from parsed PDF pages and load them into DataFrames 📈 Perform year-over-year analysis, calculate gaps between metrics, and create visualizations ⚡ Use the items view to get per-page structured data including tables and figures We demonstrate this using a 2024 Executive Summary PDF, extracting a fiscal year chart showing Budget Deficit vs Net Operating Cost data spanning 2020-2024, and reproducing the key financial insights. Check out the full tutorial: https://developers.llamaindex.ai/python/cloud/llamaparse/tutorials/parse_charts_pandas/?utm_source=socials&utm_medium=li_social Original tweet: https://x.com/llama_index/status/2027429029834531120
We’ve made an in-depth tutorial on building a PE deal-sourcing 🤝 document workflow, by simply typing a natural language prompt There’s often a set of documents coming inbound to let analysts understand what deals to invest in - teasers, CIMs, emails, presentations With our LlamaAgents builder, you can describe what documents are coming in, and the information you want to extract from each document. We perform high-quality document OCR to extract out all the key details from these documents, and give you back an application where you can process these docs at scale. Thanks to @itsclelia for this tutorial. If you’re interested, come check it out: https://cloud.llamaindex.ai/?utm_campaign=parse&utm_medium=jl_socials If you have these applications at scale, come talk to us: https://www.llamaindex.ai/contact?utm_campaign=parse&utm_medium=jl_socials
Build a private equity deal sourcing agent that automatically classifies investment opportunities and extracts key financial metrics using our LlamaAgents Builder. This step-by-step guide shows you how to create an agent that processes deal files like teasers and financial
RT Simon Suo should we rebrand to SF model harness company Original tweet: https://x.com/disiok/status/2027073086055764035
The Model Harness is Everything We are already living in a world of incredible frontier models and incredible agent tools (Claude Code, OpenClaw). But the biggest barrier to getting value from AI is your own ability to context and workflow engineer the models. This is
The Model Harness is Everything We are already living in a world of incredible frontier models and incredible agent tools (Claude Code, OpenClaw). But the biggest barrier to getting value from AI is your own ability to context and workflow engineer the models. This is *especially* true the more horizontal the tool that you’re using. If you’re using a very generic tool like ChatGPT and Claude Code, you need to spend a lot of work clearly articulating your requirements and specifications so that the agent can actually solve the task relative to your specifications. Today that looks like being extremely thoughtful about the tools that you select, and writing English very precisely in a http://skills.md file to articulate the agent these requirements. Some of the work around defining the business workflow is inherently time consuming. Think about any document SOP - simply writing the English can take hours to refine, iterate, and optimize. This is where more vertically focused agents come in; they handle the burden of equipping the agents with relevant prompts to solve a given workflow, so that you can just go in and use the application directly. Another approach is to be specialized services that offer *context* to these agents. This is the space that we (@llama_index) are operating in. We are providing the infrastructure to parse the most complex documents into agent-ready context. For other companies it could be offering web data, sales data, documentation, or codebases as a service. At a high-level any AI startup should provide context or workflows on top of these agents. We’re excited about building enduring tech even as the agent landscape evolves. If you’re specifically excited to unlock the billions of context stored within your documents, come talk to us! https://www.llamaindex.ai/contact
OmniDocBench is getting saturated VLMs are getting increasingly better at document understanding, from OSS (DeepSeek-OCR2, GLM-OCR), to frontier (Gemini 3, Kimi 5.2, GPT-5.2). A popular benchmark to measure document understanding progress has been OmniDocBench. But we're quickly approaching the point where we need a new benchmark. 1. The latest models are pushing ~95% on OmniDocBench and are already overfitting the benchmark, while still having real gaps on document capabilities. 2. The evaluation metrics of OmniDocBench depend completely on exact match and not on semantic correctness. The latter is much more important, especially in today's world where LLMs can reason over text tokens regardless of non-important formatting differences. *(see example below; our own service llamaparse is penalized even though our parsing is 100% semantically correct on scientific notation parsing)* There are so many real-world documents that haven't yet been solved by even the latest models. We'd love to welcome discussion on advancing document understanding benchmarks that are more diverse and properly score models on semantic correctness. Blog: https://www.llamaindex.ai/blog/omnidocbench-is-saturated-what-s-next-for-ocr-benchmarks?utm_source=socials&utm_medium=li_social
Document OCR benchmarks are hitting a ceiling - and that's a problem for real-world AI applications. Our latest analysis reveals why OmniDocBench, the go-to standard for document parsing evaluation, is becoming inadequate as models like GLM-OCR @Zai_org achieve 94.6% accuracy
RT LlamaIndex 🦙 Document OCR benchmarks are hitting a ceiling - and that's a problem for real-world AI applications. Our latest analysis reveals why OmniDocBench, the go-to standard for document parsing evaluation, is becoming inadequate as models like GLM-OCR @Zai_org achieve 94.6% accuracy while still failing on complex real-world documents. 📊 Models are saturating OmniDocBench scores but still struggle with complex financial reports, legal filings, and domain-specific documents 🎯 Rigid exact-match evaluation penalizes semantically correct outputs that differ in formatting (HTML vs markdown, spacing, etc.) ⚡ AI agents need semantic correctness, not perfect formatting matches - current benchmarks miss this critical distinction 🔬 The benchmark's 1,355 pages can't capture the full complexity of production document processing needs The document parsing challenge isn't solved just because benchmark scores look impressive. We need evaluation methods that reward semantic understanding over exact formatting, especially as AI agents become the primary consumers of parsed content. We're building parsing models focused on semantic correctness for complex visual documents. If you're scaling OCR workloads in production, LlamaParse handles the edge cases that benchmarks miss. Read our full analysis: https://www.llamaindex.ai/blog/omnidocbench-is-saturated-what-s-next-for-ocr-benchmarks?utm_source=socials&utm_medium=li_social Original tweet: https://x.com/llama_index/status/2026342120236396844
RT Clelia Bertelli (🦙/acc) Hey Rustacean friends!🦀 This Friday I'll be talking about how we migrated to Qdrant Edge for the storage layer of Semtools, @llama_index's Rust-powered CLI toolkit for local document intelligence🦙 The @qdrant_engine team and I will discuss lessons learnt when migrating to Qdrant Edge as a backend for on-disk vector storage in Rust, as well as the performance gains and common pitfalls to avoid👩💻 📍The live will start at 4PM CET and you can join us here: https://discord.gg/PcgpnXa2?event=1473577301470875820 If you're building AI systems in Rust, don't miss out!🦀 Original tweet: https://x.com/itsclelia/status/2026227327370469457
it's always fun when twitter hype makes a dent in the real world tbh apple should've just bought them to capture the mindshare of personalized digital assistants. 🍎🦞
Bought a new Mac mini to properly tinker with claws over the weekend. The apple store person told me they are selling like hotcakes and everyone is confused :) I'm definitely a bit sus'd to run OpenClaw specifically - giving my private data/keys to 400K lines of vibe coded
View quoted postThe second highest category is backoffice automation, but imo it's underrated by the AI community. RPA is truly dead, and agentic workflows are taking its place. A lot of backoffice work depends on routine operations over unstructured documents (invoices, claims packets, loan files). The best interface to automate these operations is enabling users to create deterministic workflows at scale, instead of solving ad-hoc tasks through chat. We are starting to build an agentic layer within our own document processing product, LlamaCloud, that lets users "vibe-code" these workflows through natural language. Come check it out: https://cloud.llamaindex.ai/
Software engineering makes up ~50% of agentic tool calls on our API, but we see emerging use in other industries. As the frontier of risk and autonomy expands, post-deployment monitoring becomes essential. We encourage other model developers to extend this research.
Agentic extraction from SCOTUS opinions 🧑⚖️ Today SCOTUS struck down Trump’s tariffs in a 6-3 vote, and one of the most interesting points was Gorsuch’s concurrence, where he devotes a decent portion calling out every justice by name - especially around everyone in the dissent (Kavanaugh, Thomas, Alito by extension). I was curious about this, so I used LlamaExtract on the official court opinion to generate a schema that not only gives me an overall summary, but extracts particular disagreements Gorsuch has in each line item. * We extract the specific disagreements with Kavanaugh and Thomas line by line * For each key, you can also trace back via bounding boxes to the source document so you can read the text! Brief: https://www.supremecourt.gov/opinions/25pdf/24-1287_4gcj.pdf Whether it’s court opinions, legal briefs, or any other complex doc, if you need to do document extraction at scale come check out LlamaCloud (particularly our “extract” feature): https://cloud.llamaindex.ai/
Coding agents are fundamentally changing software engineering in terms of velocity, role, and org structure. We published a memo to our internal engineering team detailing our growing expectations in terms of role/scope. 🟠 Before, the tasks of prioritization, engineering planning, and implementation were divided between EMs, PMs, senior ICs, and junior ICs 🟢 Now, ICs are expected to handle *all* of product prioritization, product speccing, and implementation This is due to a few trends 📈: - Coding agents have brought implementation costs down to ~0. The role of engineers is writing prompts - LLMs and sub-agents have reduced the PM work of synthesizing feedback down to ~0 too The main job of any “engineer” is to be an e2e product owner: being able to translate requirements into specifications, and delegate tasks to various subagents for implementation. Every engineer is told to offload as much as possible to their favorite tools, whether it’s Claude Code, Cursor, Devin, Codex, regular ChatGPT and more. We celebrate and share learnings around burning tokens, as long as it helps drive additional productivity!
Increased thinking doesn't correlate with increased document understanding Models are getting much better at reasoning which translates well for math/coding/general intelligence tasks. We did some experiments with GPT-5.2 on different thinking modes to see if it actually helps improves scores on OmniDocBench. Result: it doesn't, and in fact the higher the thinking, the more the output structure might deviate from the existing input. Shoutout to Boyang on the @llama_index team for this blog. Come check out our post! https://www.llamaindex.ai/blog/the-cost-of-overthinking-why-reasoning-models-fail-at-document-parsing
More reasoning doesn't always mean better results - especially for document parsing. We tested GPT-5.2 at four reasoning levels on complex documents and found that higher reasoning actually hurt performance while dramatically increasing costs and latency. 🧠 Reasoning models
RT LlamaIndex 🦙 More reasoning doesn't always mean better results - especially for document parsing. We tested GPT-5.2 at four reasoning levels on complex documents and found that higher reasoning actually hurt performance while dramatically increasing costs and latency. 🧠 Reasoning models hallucinate content that isn't there, filling in "missing" table cells with inferred values 📊 They split single tables into multiple sections by overthinking structural boundaries ⚡ Processing time increased 5x with xHigh reasoning (241s vs 47s) while accuracy stayed flat at ~0.79 💰 Our LlamaParse Agentic outperformed all reasoning levels at 18x lower cost and 13x faster speed You can't reason past what you can't see. Vision encoders lose pixel-level information before reasoning even starts, and no amount of thinking tokens can recover that lost detail. Our solution uses a pipeline approach - specialized OCR extracts text at native resolution, then LLMs structure what's already been accurately read. Each component plays to its strengths instead of forcing one model to handle everything. Read the full analysis: https://www.llamaindex.ai/blog/the-cost-of-overthinking-why-reasoning-models-fail-at-document-parsing?utm_source=socials&utm_medium=li_social Original tweet: https://x.com/llama_index/status/2024529937462706517
A lot of documents are extremely dense and repetitive in information: stapled together resumes, invoices, insurance claims, loan applications 📑 One of the main promises of AI is being able to automatically extract structured information from massive amounts of unstructured context. But in order to do this accurately over these massive documents, you need: ✅ Page attribution ✅ Bounding boxes linking back to the source text ✅ No hallucinations or dropped outputs, even if there are *hundreds* of structured fields ✅ Calibrated confidence scores to map to human review. This is exactly what we're building with our Extract feature in LlamaCloud. Our page-level extraction lets you extract massive amounts of dense information at scale from all these documents with all the auditability and citations you need, letting humans review documents much faster than before. https://www.llamaindex.ai/blog/beyond-full-text-extraction-why-page-level-granularity-matters
"It's somewhere in the PDF" is not a citation. Page-level extraction in LlamaExtract gives you: ✓ Data mapped to specific pages ✓ Bounding boxes showing exact locations ✓ Audit-ready citations Turn 200-page docs into skimmable, structured insights 👇 https://www.llamaindex.ai/blog/beyond-full-text-extraction-why-page-level-granularity-matters
View quoted postRT Ankur Goyal We sent this note to our customers to let them know that Braintrust has raised a new round of funding, and thank them for their support. While the money is exciting, our focus hasn't changed: we're building Braintrust to help our customers ship quality AI products. In 2026, AI is moving to production but teams have never had less conviction about what will fail next. Our customers are building AI products that serve millions and simply need to work. If Braintrust makes their lives easier and their products better, I know we are doing our job. Thank you to @ICONIQCapital for leading our Series B, and to @a16z, @GreylockVC, @basecasevc, and @eladgil for doubling down. Thank you to the Braintrust team for all the incredible work you've done over the past year. And thank you to our customers, who have made this growth possible. Original tweet: https://x.com/ankrgyl/status/2023810273598128588
I parsed OpenAI’s tax filings for fun 📋 I used the “Extract” capability in LlamaCloud to automatically extract out all relevant fields into structured JSON outputs. You can see their mission statement “ensure that artificial general intelligence benefits all of humanity” with the corresponding bounding boxes. (Inspired by @simonw’s post on this recently: https://simonwillison.net/2026/Feb/13/openai-mission-statement/) The extraction is powered by our core OCR engine, LlamaParse. LlamaParse is able to reconstruct this complex form PDF into markdown tables that captures each cell with ~100% accuracy. Check it out: https://cloud.llamaindex.ai/
RT ODSC (Open Data Science Conference) AI In this episode of the ODSC Ai X Podcast, we speak with @jerryjliu0, @llama_index, about the shift from early “RAG frameworks” to document AI workflows. 🎧Listen to the full episode here - https://hubs.li/Q042_7Wq0 Original tweet: https://x.com/_odsc/status/2022838519886926130
Existing AI agents are largely short-horizon (e.g. chat) or constrained (e.g. agentic process automation). @sequoia predicted that 2026 is the year of long-horizon agents, and long-horizon agents ~= AGI We’re already seeing this with coding agents. But I’m specifically excited about how this translates to non-technical knowledge work over documents (e.g. legal, finance, insurance, back office) that humans typically perform for hours on end. I wrote a blog post that articulates the broader class of use cases this would unlock, and the ideal architecture/UX of an “agent inbox” that allows agents to run autonomously without being blocked on human input. Unlike a chat interface. Some examples: 1. An agent that can continuously create a living FAQ from your Sharepoint, Slack, call transcripts 2. An agent that can e2e interface with lawyers to do contract redlining This fundamentally requires building agents that can monitor event triggers beyond human chat inputs, and can use core capabilities around parsing, extracting, creating, and editing documents of various formats. Check out the blog! https://www.llamaindex.ai/blog/long-horizon-document-agents If you’re interested in the building blocks around parsing, extraction, with more coming soon, come check out LlamaCloud: https://cloud.llamaindex.ai/
RT LlamaIndex 🦙 2026 is the year of long-horizon agents. @sequoia predicts that this year, agents will be able to tackle long-horizon tasks and work autonomously for hours to solve ambiguous tasks. We're excited about how this translates to knowledge work automation, particularly over documents. Let's take a look at "Long Horizon Document Agents" 🕰️ Agents are evolving to work autonomously over weeks, not just minutes, handling complex document tasks end-to-end. 🔄 These agents can continuously monitor events like document changes, comments, and deadlines - not just respond to chat prompts 📝 They maintain persistent task backlogs and can collaborate iteratively on living documents like FAQs, PRDs, and legal contracts 🎯 The interface shifts from chat boxes to "agent inboxes" that manage ongoing document tasks with clear status and context ⚡ This enables true automation of multi-step knowledge work - from due diligence memo updates to contract redline collaboration loops 2026 is shaping up to be the year agents evolve from "workflows" to "employees" - and we're building the document processing infrastructure to make this possible. Read @jerryjliu0's full blog on long horizon document agents: https://www.llamaindex.ai/blog/long-horizon-document-agents?utm_source=socials&utm_medium=li_social Original tweet: https://x.com/llama_index/status/2021992214038241477
We love parsing diagrams. Anthropic’s recent report on coding trends has a nice diagram on the evolution from single-agent to hierarchical multi-agent architectures With our latest VLM-enabled document parsing, we’re able to one-shot this diagram into a `mermaid` plaintext representation! Check out the results below. This capability lets you convert even the most complex diagrams within PDFs/Powerpoints into digestible graph representations that LLMs can understand. This lets you use AI to understand complex docs at scale; VLMs either can’t understand these diagrams out of the box, or you also end up burning unnecessary vision tokens. The report itself is an interesting overview of multi-agents, check it out: https://resources.anthropic.com/hubfs/2026%20Agentic%20Coding%20Trends%20Report.pdf?hsLang=en For diagram parsing, sign up to LlamaCloud: https://cloud.llamaindex.ai/ If you’re interested in chatting more about this, come talk to us: https://www.llamaindex.ai/contact
who needs Clay for outbound personalization when you have this
Coding agents are not great for open-source software, and it needs to adapt This is a really nice article from @LoganMarkewich that draws from our experience maintaining our own OSS library in the past 3 years. 1. Open-source used to be nice for human knowledge sharing. Now you really need to optimize it so that coding agents can pick it up. 2. Light abstractions (wrappers around LLMs, vector databases, etc.) are dead 3. There is a massive slopocalypse of ai-generated PRs, which breaks the fun community feeling of OSS. If you want to lean in and accept this, you need to document patterns clearly so coding agents can pick it up https://www.llamaindex.ai/blog/on-the-incoming-slopocalypse-and-the-death-of-open-source?utm_source=socials&utm_medium=li_social
The rise of coding agents is fundamentally changing open source - Our head of OSS @LoganMarkewich breaks down how LLM-powered coding agents are impacting core pillars of open source: 👥 Community interaction, which is getting complicated by low-quality, massive AI-generated PRs
RT LlamaIndex 🦙 The rise of coding agents is fundamentally changing open source - Our head of OSS @LoganMarkewich breaks down how LLM-powered coding agents are impacting core pillars of open source: 👥 Community interaction, which is getting complicated by low-quality, massive AI-generated PRs 💪 Personal skill development suffers when developers rely too heavily on AI assistance 🧠 Knowledge sharing is shifting as LLMs become the frontend for learning But open source isn't dead - it's evolving. We're shifting toward hackable reference implementations, community-driven knowledge sharing, and agent-friendly codebases that work with AI tools rather than against them. Read the full blog by Logan on how he views this evolution of open source projects: https://www.llamaindex.ai/blog/on-the-incoming-slopocalypse-and-the-death-of-open-source?utm_source=socials&utm_medium=li_social Original tweet: https://x.com/llama_index/status/2021631328236802318
Scaling Document Ingestion for AI Agents We're excited to partner with StackAI on this webinar. We'll show you how to build a modern agent stack that can automate knowledge work over millions of documents, across finance, legal, insurance use cases and more. Come check it out! https://lnkd.in/gzrWt33y
Are you trying to solve high-quality document ingestion for your product? Gain lessons from the field on how @stackai uses LlamaCloud to power high-accuracy document ingestion & retrieval across PDFs, images, spreadsheets & more — at enterprise scale. ➡️ Register now:
RT LlamaIndex 🦙 Are you trying to solve high-quality document ingestion for your product? Gain lessons from the field on how @stackai uses LlamaCloud to power high-accuracy document ingestion & retrieval across PDFs, images, spreadsheets & more — at enterprise scale. ➡️ Register now: https://landing.llamaindex.ai/webinar-stackai-and-llamaindex-scaling-document-ingestion Original tweet: https://x.com/llama_index/status/2021259881823834552
RT ODSC (Open Data Science Conference) AI In this episode of the ODSC Ai X Podcast, we speak with @jerryjliu0, @llama_index, about the shift from early “RAG frameworks” to document AI workflows. 🎧Listen to the full episode here - https://hubs.li/Q042hjhv0 Original tweet: https://x.com/_odsc/status/2021026586779111700
Yes he is on Singles Inferno but also we would unironically hire him (Samuel if you can’t find love you can always find AI)
In hindsight, a quant trader going on a Netflix dating show was definitely the top signal…
Parsing line charts is a hard task for VLMs VLMs are generally fine at coarse visual understanding, but they have a hard time reasoning about precise coordinates. Ask most VLMs, even though tuned to chart understanding, to parse a line chart to a table and they will struggle. We tested over a few samples. Docling’s new granite-vision model, gemini 3 flash, gpt 5.2 pro, and a v0.1 of our own chart parsing (which is in beta and rapidly evolving). Out of these, most models fail, and sometimes miss the entire chart correctly. gpt 5.2 pro is closest but spends an absurd number of tokens reasoning through each point. Our own parsing is actually quite good, though of course, there’s still some things we need to do to get to 100% accuracy. If you want to parse complex documents with diagrams/charts, come check out LlamaCloud! https://cloud.llamaindex.ai/
Introducing Claude Opus 4.6. Our smartest model got an upgrade. Opus 4.6 plans more carefully, sustains agentic tasks for longer, operates reliably in massive codebases, and catches its own mistakes. It’s also our first Opus-class model with 1M token context in beta.
View quoted postRT LlamaIndex 🦙 Ready to master production-grade multi-agent AI systems in one intensive day? 🚀 We're partnering with @AWS and leading AI companies for the AWS AI Builder Lab with @aicampai in San Francisco on Feb 13th - a hands-on competition where you'll build sophisticated agentic workflows, not basic chatbots. 🤖 Design and orchestrate powerful multi-agent workflows using cutting-edge tools 🏆 Compete in real-time challenges with live leaderboards and prizes for top performers 🔧 Get production-ready patterns you can implement immediately, plus enterprise tool trials Coordinate intelligent systems to tackle real-world challenges. Limited to 250 qualified developers. Register here: https://www.aicamp.ai/event/eventdetails/W2026021308 Original tweet: https://x.com/llama_index/status/2019139663462850628
Our parsing models are able to parse this massive diagram into mermaid format with 100% accuracy. Original doc on the left, parsed mermaid on the right. This is the “agentic plus” mode in LlamaParse, powered by state-of-the-art VLMs / agentic reasoning that can interpret complex relationships within any document page. If you have docs with a lot of flowcharts and diagrams, come check us out! https://cloud.llamaindex.ai/
RT Tuana MCP connects agents to live systems: databases, APIs, external services. It's designed for runtime tool access . But the moment you need to teach your agent how to approach a problem domain, you need something else. Skills aren't about (just) accessing data. They're about embedding knowledge into your agent's reasoning. When your agent needs to understand "here's the right sequence for debugging a data pipeline" or "this is how you validate and process complex documents," skills allow you to bake that knowledge into how the agent thinks. There's then also the whole matter of how they work fundamentally: MCP tools rely on an external connection and API calls. Skills are local.. The issue isn't choosing between them. It's understanding that they kiiinda serve different purposes. MCP extends your agent's capabilities at runtime. Skills shape how your agent reasons about problems. I wrote all about this with @itsclelia in our latest blog: https://www.llamaindex.ai/blog/skills-vs-mcp-tools-for-agents-when-to-use-what Original tweet: https://x.com/tuanacelik/status/2019106807437038029
no one's mentioned pie punks yet pie punks is extremely underrated
we spent a lot of time figuring out how to context engineer claude code with mcp tools and skills this comparison diagram should help you understand the tradeoffs between the two 👇
Are you choosing between MCP servers and skills for your agent? @tuanacelik and I wrote a blog post about the differences between the two and their PROs and CONs, informed by the lessons learnt while building our own coding agent (LlamaAgents Builder). The main takeaway? Skills
RT Clelia Bertelli (🦙/acc) Are you choosing between MCP servers and skills for your agent? @tuanacelik and I wrote a blog post about the differences between the two and their PROs and CONs, informed by the lessons learnt while building our own coding agent (LlamaAgents Builder). The main takeaway? Skills are easier to set up but tougher to keep up-to-date, whereas MCPs are more dev-facing, but offer a more deterministic toolset with less maintenance overhead. Check out the article 👉 https://www.llamaindex.ai/blog/skills-vs-mcp-tools-for-agents-when-to-use-what Original tweet: https://x.com/itsclelia/status/2018821269752611102
Should you use MCP or skills in your coding agent (?) We gained a lot of insight into this when we built our own coding agent (the LlamaAgents builder) within LlamaCloud. tl;dr Skills are way easier to setup, but unreliable/hard to maintain. Blog: https://www.llamaindex.ai/blog/skills-vs-mcp-tools-for-agents-when-to-use-what?utm_source=socials&utm_medium=li_social
Confused about whether to use Skills or MCP tools for your AI agents? We break down the key differences and when to use each approach. 🔧 MCP tools provides a more deterministic interface with fixed schemas - perfect for precise, predictable operations but require more technical
RT LlamaIndex 🦙 Confused about whether to use Skills or MCP tools for your AI agents? We break down the key differences and when to use each approach. 🔧 MCP tools provides a more deterministic interface with fixed schemas - perfect for precise, predictable operations but require more technical setup and an overload of MCP tools can cause a confusion of choice to an agent 📝 Skills use natural language instructions in markdown files - minimal setup required but open to LLM interpretation variations ⚡ MCP involves network latency while Skills run locally, but MCP offers centralized updates that propagate automatically Read our full analysis with real examples: https://www.llamaindex.ai/blog/skills-vs-mcp-tools-for-agents-when-to-use-what?utm_source=socials&utm_medium=li_social Original tweet: https://x.com/llama_index/status/2018749615907213457
A week after PaddleOCR-VL-1.5 took the top spot on OmniDocBench, *another* 0.9B model dethrones it! GLM-OCR shows SOTA results on doc parsing benchmarks and it's apparently 50-100% faster https://huggingface.co/zai-org/GLM-OCR
Introducing GLM-OCR: SOTA performance, optimized for complex document understanding. With only 0.9B parameters, GLM-OCR delivers state-of-the-art results across major document understanding benchmarks, including formula recognition, table recognition, and information extraction.
I built a simple paralegal agent to detect, classify, and extract information from court filings🧑⚖️- complaints, motions, orders, and more! Wrote a prompt in English, and it encoded it into a deterministic, repeatable workflow that you can use to run on millions of court filings at scale. 1. A new docket entry arrives 2. Detect what type of filing it is (motion/order/complaint/etc) 3. Define a separate schema for each new filing 4. Synthesize results in a UI with human review. Can also plug it in as an API into your workflows This is available to everyone in our LlamaCloud agents builder. The prompt is exactly what you see in the screenshot; you can make it more/less verbose if you’d like. Come check it out! LlamaCloud by @llama_index: https://cloud.llamaindex.ai/
