AI Research

Google Research AI document reviewer that checks technical arguments, mathematical reasoning, related literature, and cross-section consistency in papers and complex documents. Includes two free reviews per week; additional reviews use your own Gemini API key at standard model token rates.

Vela 2.0 is a family of Apache-2.0 open routing models from the vLLM Semantic Router project that answers routing, safety, PII, and groundedness questions in one request and returns span-level decisions with probabilities.

Microsoft's open-source framework for defining MCP tool environments, running LLM agents with simulated users, and evaluating final backend state, side effects, and response requirements for reliable agent workflows.

Vicuna-13B is an open-source conversational language model from LMSYS, fine-tuned from LLaMA on user-shared conversations and released with training code, model weights, and a research demo for non-commercial use.

Kolibri is Aleph Alpha's Apache-2.0 open-weight German-English mixture-of-experts reasoning model with 78B total and 3.46B active parameters, up to 1M-token context, tool calling, and self-hosted vLLM deployment.

AI-generated online encyclopedia from xAI that creates and continuously updates readable knowledge articles, checks proposed edits against sources, and lets readers suggest topics and corrections.

Open-source local AI agent inference benchmark from Artificial Analysis. It replays recorded agent trajectories against OpenAI-compatible model servers to measure throughput and latency on laptops and workstations.

Open-source, local-first document parsing library and CLI that extracts PDFs, Office files, and images into spatially grounded text and Markdown, with OCR, bounding boxes, document-complexity detection, and Node.js, Python, Rust, and WASM support.

Toast 1 is Mixedbread's agentic search model: it decomposes a query into steps, runs parallel retrievals with "next file search" iteration, inspects sources, and curates evidence before returning. It runs standalone or as a retrieval subagent, delivering frontier search quality at roughly 1/10th the cost and 12x the speed of comparable frontier models.

anydoc is Firecrawl's open-source document-to-Markdown converter that turns Word, PowerPoint, Excel, PDF, OpenDocument, RTF, EPUB, and CSV files into clean GitHub-Flavored Markdown in milliseconds. It runs fully local via CLI, Node.js, Python, Rust, or WebAssembly and ships as an Agent Skill so coding agents can read any document.

Local web-intelligence layer for AI agents that searches, fetches, crawls, extracts, caches, and finds similar pages through MCP, REST, SDK, and framework integrations.

Open-source enterprise digital employee platform for turning SOPs and documents into traceable AI workflows with scheduled execution, MCP and HTTP tools, memory, and human takeover.

Long-term memory runtime for AI agents that combines Markdown-native memory, hybrid retrieval, and self-evolving skills so agents can retain context and improve across sessions.

Enterprise document intelligence model from Mistral AI with 170-language OCR, bounding boxes, block classification, and self-hosted deployment for large-scale extraction workflows.

OpenAI-compatible multi-agent model interface that coordinates specialized AI agents behind one API, routing complex tasks through a unified Fugu and Fugu Ultra workflow.

Google AI research assistant grounded in your own sources. Upload documents, discover web sources, run code in a secure cloud computer, and generate reports, slides, and audio overviews.

WhatIsThisMovie is a movie discovery tool that helps users identify films from descriptions, scene clues, or simple keywords. In seconds, users can find the title, release year, cast, genre, and related recommendations.

Shared long-term memory layer for AI agents that turns user preferences, facts, profiles, and intent into reusable structured memory across sessions and agent workflows.

No-code AI platform for predictive analytics — build, deploy, and explain machine learning models without writing code.

Industry-leading open-source OCR toolkit supporting 80+ languages with document parsing, table recognition, and Transformers backend.

Open-source AI assistant for Datasette that answers data questions via conversational SQL queries against SQLite databases.

AI-native search API designed for AI agents and RAG applications — delivers clean, structured search results optimized for LLM consumption.

AI search engine with personalized answers, research mode, coding assistance, and image generation — all in one search interface.

AI agent memory layer built by YC CEO Garry Tan — markdown-first knowledge system with auto-wiring knowledge graph, semantic search, and MCP integration for persistent agent memory.