LlamaIndex Newsletter - Looking Back on 2025 - LlamaIndex

Welcome to our special 2025 Year-End Edition! As we wrap up an incredible year of building the future of document AI and prepare to ring in 2026, we're feeling extra grateful for this amazing community. This year has been absolutely wild - from launching game-changing tools like LlamaAgents, LlamaSplit, and LlamaSheets to all manner of MCP support across the board. We've seen you build everything from financial document processors to coding agents that can safely explore entire codebases. As we toast to 2025 and look ahead to an even more exciting 2026, here's a celebration of our biggest wins, coolest demos, and the incredible features you've built with us. Cheers to another year of making AI work in the real-world! 🥂 LlamaAgents Launch: We launched our one-click document agent deployment product based on our open-source Workflows library, with ready-to-use templates for invoice processing, contract review, and claims handling. Learn more here Document AI Revolution: We've moved beyond traditional OCR to create intelligent systems that truly understand documents like humans do, achieving 90%+ pass-through rates vs 60-70% with legacy systems through agentic workflows and multimodal understanding. Blog post Context Engineering: This topic was highly discussed in 2025 as a new approach to thinking about how we provide the right context to LLMs so that they resolve queries and complete tasks in the most efficient way. Read our blog on this topic Parse vs. Extract Deep Dive: Our comprehensive technical guide that became the go-to resource for understanding when to parse documents into structured markdown versus extracting specific fields with schema validation. Technical guide LlamaParse v2: Redesigned with four simple tiers (Fast, Cost Effective, Agentic, Agentic Plus) replacing complex configurations, plus up to 50% cost reduction. Announcement LlamaSheets Beta: Transform messy spreadsheets into AI-ready data with intelligent region classification, multi-stage processing, and 40+ features per cell. Get started LlamaSplit: Automatically separate bundled documents into distinct sections using AI to analyze page content and group consecutive pages by category. Try beta LlamaExtract Table Row Mode: Extract data from repeating entities like catalogs and tables with precision, solving the core problem of losing 80% of data when processing long documents. Learn more Workflow Debugger: Shipped with built-in observability features allowing you to visualize workflows, see event logs in real time, and compare runs. Docs Agent Templates: Created a collection of pre-built templates through llamactl covering document Q&A, invoice processing, and data extraction with one-command deployment. Browse templates Safe Coding Agents: Built AgentFS integration for secure filesystem access, allowing coding agents to work with virtual filesystems without damaging your real files. GitHub TypeScript Workflows: Launched comprehensive TypeScript support with Express agent tutorials and production-ready deployment patterns. Get started SemTools CLI: Added dedicated "ask" command for agentic search over documents, combining with parse to create QA workflows over unstructured data. Learn more Open-Source NotebookLM Alternative: The community built a complete document-powered AI app architecture with event-driven workflows, automated document ingestion, and state management - showing the full power of our platform in action. GitHub Holiday Fun: We built agents to help Santa process corrupted toy databases and extract wish-list items from thousands of letters - proving AI can make even the North Pole more efficient! Santa's helper Colab Customer Success Stories: Pathwork scaled from 5,000 to 40,000 pages per week, Intelligence Co. built Cofounder AI chief of staff, and MavenBio transformed scientific visuals into searchable intelligence. Case studies Filesystem Explorer Agent: Built with Gemini 3 Flash for real-time file exploration with tool-powered navigation and interactive questioning. GitHub StudyLlama: Community-built web app using LlamaAgents to organize study materials with automatic categorization and Q&A generation. GitHub MCP Integration: Added native Model Context Protocol support for documentation search, enabling coding agents to directly access our docs. Try it at: https://developers.llamaindex.ai/mcp As we head into the new year, we're more excited than ever about what's coming. Document AI is just getting started, and with your creativity and our tools, we're going to build some incredible things together. Here's to a fantastic 2025 and an even better 2026! Keep building, keep innovating, and keep being awesome.