AI (Artificial Intelligence) Startups funded by Y Combinator (YC) in San Francisco 2025

Scale accelerates the development of AI within organizations of any size to deliver critical business insights and operational efficiency. Its data-centric infrastructure platform leverages RLHF (Reinforced Learning with Human Feedback) to help organizations build the strongest AI models that supercharge their business, with customers across industries including Meta, Microsoft, U.S. Army, DoD’s Defense Innovation Unit, Open AI, General Motors, Toyota Research Institute, Brex, Instacart and Flexport.

Checkr builds people infrastructure for the future of work. And we believe everyone should have a fair chance to work. That's why we've designed a faster—and fairer—way to screen job seekers. For more information on our mission and products, visit https://checkr.com.

Didit is an AI-native, developer-first identity verification platform that makes KYC, AML, and fraud prevention instant, flexible, and transparent, with a seamless user experience and more than 70% lower costs.

Fifth Door unlocks the power of imagination by making it easy for everyone to create, share, and play games anywhere, with anyone. We exist to stimulate the creativity, connection, and joy that people get when they play games.

AVoice brings full visibility to the physical side of sales. Most sales intelligence tools only track digital interactions — emails, calls, CRM entries — but the majority of global revenue is still driven by in-person visits. Reps take photos, record voice notes, send quick messages, and make on-site observations, but this information is scattered, unstructured, and rarely reaches the CRM. AVoice automatically captures, analyzes, and structures everything that happens during a field visit. Photos, voice notes, text updates, and observations are instantly transformed into searchable insights for reps, managers, and leadership. AVoice highlights risks, opportunities, buying signals, product issues, market trends, and territory health in real time. By turning the offline economy into usable data, AVoice gives companies the visibility they’ve always had for digital sales — now applied to the real world where deals are actually won.

AtlasGrid automatically builds a visual graph of your mobile app UI and displays analytics directly on your screens. Additionally, our fleet of AI-powered agents will continuously analyze thousands of mobile apps, gathering design insights and identifying growth opportunities + actionable product improvements.

Enterprises are racing to adopt AI, but often don't have the right controls in place or know how to verify vendors are compliant. LogosGuard turns AI policies into executable controls, stress-tests your AI products, and continuously monitors changes, giving enterprises the confidence to scale AI responsibly.

We’re building AI agents that speed up drug trials by automating the manual consolidation and document prep required for every IND submission. Today, biotechs spend months stitching together reports from contractors and internal teams just to file and get to their first-in-human trial.

Exonic is building a global network of intelligence to cure disease. We are the first drug discovery company open to everyone. Today drug discovery is siloed - secretive, adversarial institutions in glass towers. Biological AI is changing the rules. Every day drug discovery becomes more of a computational problem. Exonic is turning drug discovery into an open, meritocratic tournament. We are advised by world leaders in bio AI, including the former Head of AI at Ginkgo Bioworks, and the first author of Evo (cover of Science). All of the best biological AI models are open source. And on Exonic, you can combine them, prompt them, and share your innovative workflows without writing a single line of code. Using Exonic Studio, we just beat a state-of-the-art benchmark in liver cancer (HepG2) synthetic enhancer design, validated in our wet lab. The platform just went live - you can start right now. We are turning the internet into a drug discovery powerhouse.

Thesis is building the engine for discovery. Our aim is to help uncover the next AlphaFold or build the next Transformer, faster, more systematically, and at scale. We treat discovery not as a matter of luck, but as an engineering challenge to be solved. Thesis lets anyone build powerful machine learning models for anything they can imagine. The mission is to democratize invention by putting machine intelligence in everyone’s hands. Every major breakthrough in the next decade, in science, industry, or AI, will be built on systems like Thesis. We intend to define that category.

Castari lets developers deploy AI agents in secure, autoscaling sandboxes with one command. We handle tool/MCP connectors, snapshots, and observability so teams go from prototype to production in hours instead of weeks. Developers focus on building agents, Castari provides the autoscaling sandbox infrastructure to ensure your agents and its tools are secure.

Zarna is private equity’s first AI associate class. Our agents plug directly into your firm’s data and work like a deal team that never sleeps. Zarna’s agents analyze CIMs, build LBOs, update CRMs, draft IC memos, & act based on decades of proprietary firm knowledge. The result: investors spend less time buried in decks and spreadsheets, and more time on judgment, relationships, and winning deals.

Nessie is building personal AI brains that augment human cognition. Nessie turns your scattered AI conversations and documents into a shareable AI brain you actually own. It's a thinking partner that lets you instantly search your entire history of ideas, synthesize them into new insights, and watch your knowledge build on itself over time. Right now, your best thinking is trapped in ephemeral AI chat histories. Nessie unifies your context from ChatGPT, Claude, and (soon) Apple Notes/Google Docs/Notion/other context sources, auto-organizing the mess into an AI-enhanced brain. Next, we are unlocking sharing. Instead of sending a static doc, you will be able to send your "Product Strategy Brain" to a cofounder, or publish your "Research Brain" for your entire team to chat with. Who it's for: AI-native thinkers—founders, researchers, and writers—who believe their best ideas should compound. The Team: We are two Yale CS grads and former Amazon engineers building the tool we needed for ourselves. We combine Anna's obsession with neuroscience and augmenting human cognition with Tiger's lifelong passion for building products (he shipped an app with 70k+ users at age 13). Traction: Launched 3 weeks ago. We have over 1,200 users who have imported 300,000+ AI conversations into Nessie. 30% week-over-week organic user growth. Origin Story: Nessie was born after we hit the 1-million-token limit in multiple Gemini chats. Overnight, months of research disappeared. We realized the problem wasn't just about storage; it was that the medium for AI-native thought didn't exist yet. The chat window is a place to ask questions, not a place to build a mind. Our Vision: We are building a new medium for personal intelligence. Everyone will have an AI brain on Nessie. You will use it to think better, remember more, and share your mind with the world.

DiligenceSquared delivers AI-powered market due diligence for investment decisions - like the $500k McKinsey reports, but automated, auditable, and at a fraction of the cost. When Private Equity funds evaluate deals, they pay firms like McKinsey, BCG, and Bain $500K-$1M for market due diligence. The process takes weeks, and the output is a static PowerPoint that's impossible to audit. A single large-cap PE fund can spend >$30M annually on this and the total market is $10B per year. Our AI agents conduct expert interviews and synthesize insights with every claim linked to source transcripts. No black box, no "trust me" slides, and no 10x mark-up for human labor. Frederik bought these reports as a Principal at Blackstone. Soren sold them as a Principal at BCG. Harshil is a repeat founder and former Software Engineer at Google. We've seen both sides and know what matters in due diligence - and now we automate it. Our clients already include 8 leading private equity funds with over $2.4 trillion in combined AUM.

Jarmin.ai is your 24/7 ML eng employee that you hire. Just talk to Jarmin like any other employee and Jarmin handles your AI/ML work from there. Completely hand over full initiatives or individual tasks. Jarmin’s founding team is from Meta SuperIntelligence Labs (Staff Engineers), Apple (Machine Learning Engineer), AWS (scaling infra), Lockheed-Martin (Research), and JPMorganChase (SWE) with experience building AI/ML and automating complex work of Researchers, Data Scientists, ML Engineers with AI.

Sanctum lets you launch new features to agents that mimic your real users. We build AI user models from your session recordings and run them against your changes to surface bugs, UX issues, and performance problems early.

Hireglide is an AI screening tool for hiring teams. Instead of wading through 500 CVs, we screen them and surface the best candidates. In future, with access to millions of CVs, we will match candidates with relevant roles.

Efference builds the eyes and visual cortex for robots. Instead of treating depth as a hardware problem, we use software — inspired by how humans see — to generate rich and reliable 3D information. This lets us deliver higher performance and greater ease of use at a lower price than existing stereo cameras.

Inspector helps front-end engineers ship faster with AI by automating the most frustrating part of the job — gathering context between the browser and the codebase. Inspector is an AI IDE for front-end that connects your browser and codebase. Inspector lets you select elements visually and prompt changes, giving coding agents access to the same runtime context developers have: visuals, elements, and logs. With Inspector, you’re editing right in the front-end, not buried under tabs and screenshots.

Social networks are dead. Social media became more media than social. My friends aren't on social media anymore - it's one third ads, one third content from strangers, and one third AI slop. Our social lives are fragmented across dozens of different apps - iMessage, WhatsApp, Find My, Partiful, and Venmo. Digipals is the first AI-native social operating system that improves our relationships with friends. We built ambient AI agents that use social context from our location, calendar, and photos to take action and write code in our group chats. We are the last social platform anyone will ever need. Launch Video https://youtu.be/KW2tjLl6sxM

Sava is building the agentic trust company for an industry that manages $6.5 trillion in U.S. assets today. A trust charter is the state license that lets us act as a legal trustee: open and administer trusts, hold assets, and make distributions under the law. We treat that charter like infrastructure. Think Stripe for fiduciary work: one integration and dashboard to open a trust, fund it, track it in real time, make distributions, and stay compliant. Our AI agents handle the grunt work: collect data, chase signatures, assemble documents, reconcile positions so that people can focus on making high-quality decisions. Every step is captured in a timestamped audit trail. Agents also enable continuous compliance: risk checks run as data changes. Reviews are prompted automatically. Nothing falls through the cracks. Team Nimit Maru, CEO – 2X YC Founder. Built and sold Fullstack (S12) for $55M on $1M raised. Experienced the trust industry's dysfunction firsthand post-exit and came back to YC to fix it. Rush Sadiwala, COO – Experienced founder, previously backed by Kevin Rose at True Ventures. Spent the first 5 years of his career on the buy side in Investment Management. We're in YC F25 and working with experienced trust professionals, including a Chief Trust Officer from a top-tier trust company, to bring modern infrastructure to the trust industry.

Sylvian creates expert data for LLMs through competitions. LLMs continue to need lots of expert data. Unfortunately, existing data vendors like Scale or Mercor are not motivating the best experts through part-time pay. Competitions are more intrinsic motivators, and naturally incentivize high quality data. Sylvian creates expert data for LLMs through competitions, starting with tool use data (e.g. Excel, VSCode). We already have 4,500+ experts, from IMO Golds to MIT/Stanford PhDs to full time QRs at Point72, producing data at 1B tokens/week!

Anto is building a foundation model for microbial communities, making the gut microbiome computable for the first time. We predict drug toxicity and efficacy across diverse populations and fix drugs so they work for everyone — solving the hidden root cause of most drug failures. Founded by Arvid (Broad Institute of MIT and Harvard; Nature-published researcher who pioneered quality-aware, goal-directed sparsification) and David (Harvard Medical School Gastroenterology, J&J), second-time founders who published the breakthrough at leading machine learning AI conferences and Nature.

Norra helps nursing facilities manage equipment operations. Every year, ~$80 billion are spent on medical equipment like beds, concentrators, and wheelchairs. Nursing facilities continue to operate with zero visibility which leads to $15 billion in equipment waste annually. We use physical trackers and AI agents to eliminate this.

Lemma enables AI agents to continuously improve from user feedback and production outcomes. Instead of relying on manual prompt tuning and reactive debugging, Lemma detects when an agent’s performance drifts, pinpoints the exact step causing failures, generates optimized prompt candidates, and automatically delivers improvements through our API or by opening a pull request in your codebase. Built by ex-AI engineers who felt this pain firsthand, Lemma turns static agents into self-improving systems, helping teams ship reliable AI products faster with compounding performance gains over time. Deploy 1x. Learn forever.

Rovr streamlines engineering operations by automating intake, triage, routing, and cross-team coordination. It replaces internal forms, manual Slack triage, and scattered docs with an AI front desk that understands requests, assigns owners, updates context, and posts status changes directly into Slack. Rovr’s built-in ops CRM gives teams a single system of record for eng requests — from tooling access to platform incidents to dev-infra workflows — reducing operational drag across the entire org.

Logical is a desktop AI that acts as a proactive and largely promptless copilot, designed to reduce the friction of using AI for everyday tasks. It transfers context across apps, and allows taking action on many apps with just a few clicks. Logical can assist with drafting emails, writing docs, noting meetings and can even automatically detect action items. Need to prompt? Logical can understand intent in much fewer words than a regular chatbot. Plus, user data is never sent to Logical servers. Logical is the AI resurrection of Clippy, but actually good.

Moss is building the real-time semantic search runtime for conversational and multimodal AI. Our system enables voice agents, copilots, and chat interfaces to retrieve, reason, and respond in sub-10 ms, delivering the responsiveness that makes AI interactions feel truly natural. If you’ve ever built a conversational or voice AI product, you’ve felt the lag. That moment when an agent pauses and the illusion of intelligence breaks. The bottleneck is almost always retrieval. Each query hops across networks and databases, adding delay and cost. Moss eliminates that gap by keeping retrieval close to where the agent runs. Moss runs natively across browsers, mobile devices, and servers with an optimized vector index built in Rust and WebAssembly. It enables teams to build AI products that feel instant, contextual, and adaptive. The experiences that sustain engagement and unlock new kinds of interaction. For our customers, this translates into tangible business value: stronger user retention, higher conversion, and entirely new product categories made possible by real-time understanding. Moss is already powering production pilots across voice AI and developer platforms, achieving sub-10 ms retrieval and 70–90% token savings compared to traditional pipelines.

PlayVision is the AI Moneyball for Sports: the future for scouting, recruiting, and playing the game. Basketball teams upload their practices or games, and PlayVision uses AI and computer vision to automatically tag and analyze the film, giving coaches and scouts 10 times the information on their players and their opponents. PlayVision also uses algorithms to rank every single college and professional basketball player in the country with over 40 different metrics from tracking and play-by-play data. Check out our website at withplayvision.ai, and follow our instagram @playvisionai!

An AI companion that chats and plays with children inside games like Minecraft, helping them learn a new language while they play. Kids already spend hours each week in their favorite games, and now parents can turn that game time into learning time.

Mantle connects to all your tools: your CRM, Mail, Calendar, Payments, Product Analytics, and more. With just one prompt, you can create agents that handle back office work for you. These agents can automate reports and email them to you, drafts emails right in your email client, generates pre-meeting research directly in your calendar event, and much more. They run quietly in the background, react to changes in real time, follow a schedule or work when you chat with them. You didn’t start up to stay stuck in admin. Turn your company AI-native with Mantle.

Arcten enables companies to embed AI copilots directly into their applications so users can accomplish complex workflows through natural language. You can add AI copilots to your product in minutes, connect them to your company’s information, and track how they’re performing -- all in one place.

Velvet is the all-in-one tool for making AI videos. Our team has worked on video generation at Meta FAIR and infra at Adobe, and we’re building the single tool that companies need to produce AI video assets.

Private markets are still opaque, manual, and locked inside Excel. Every year, trillions of dollars in venture, real-estate, and private-equity capital are deployed through diligence processes that haven’t changed in decades. Analysts spend hours parsing PDFs and spreadsheets just to understand what a company or property is worth. Alt-X is building the data and AI infrastructure layer for alternative assets — a system that ingests unstructured deal materials, normalizes the data, and generates institutional-grade analysis in seconds. Whether it’s a private-company dataroom or a real-estate offering memorandum, Alt-X extracts key KPIs like ARR, growth, NOI, IRR, and cap rate, then produces investor-ready dashboards and reports automatically. Our platform reduces deal evaluation timelines by up to 73%, saves 30–50 analyst hours per deal, and enables teams to screen 4× more opportunities with the same headcount. By automating KPI extraction, benchmarking, and underwriting, Alt-X transforms diligence from a manual bottleneck into a scalable process. This hasn’t been done at scale because private-market data is fragmented across formats and asset classes. Alt-X’s parsing engine unifies this chaos into a structured schema — the backbone for cross-asset underwriting, KPI dashboards, and automated risk scoring. Our first wedge is real estate and private equity, where underwriting is slowest and most manual. We’re already piloting with family offices and private funds managing over $200 million in assets, paying $1K–$5K per month to automate deal screening and valuation. The opportunity is massive: over $1.4 trillion in new private-equity and real-estate deals are underwritten globally each year, with $25–30 billion spent on diligence and analytics. Yet over 90% of this still happens manually. Alt-X’s mission is to become the operating system for private-market investing — the Bloomberg Terminal of the AI era.

We are building Siri for work, but it actually works. Caddy turns your voice into the next computing interface. It understands what’s on your screen and your intent, so work gets done while you stay in flow. Say “create a Linear ticket and share it to Slack” and Caddy handles it instantly in the background. Caddy’s founders previously led product and design for Loom’s AI suite and core products. We believe the next era of computing will be powered by voice, where computers understand what you say and what’s on your screen so work happens naturally.

Nivara gives engineering leaders visibility into how humans and AI build together, helping teams spot bottlenecks, measure impact, and turn AI adoption into real outcomes. We're building the much needed copilot for Engineering Managers and empower every team to work intelligently with AI.

Building the data infrastructure for robots. Large-scale, real-world data is the missing ingredient for general-purpose embodied agents.

Boom builds the AI employee for e-commerce — an agent that manages every customer interaction across WhatsApp, Instagram, TikTok, Facebook, and web chat. It answers product questions, supports customers with orders, follows up on abandoned checkouts, restocks, and repeat purchases, and even runs marketing campaigns. Brands simply connect Boom to their store, and their AI agent instantly starts selling, supporting, and following up like a real team member — no setup or flow-building required. E-commerce is a $6T market where brands spend heavily to bring customers in, only to lose over 99% of them before checkout. Boom fixes that by turning every missed conversation into a sales opportunity. Our clients increase conversion rates and deliver a concierge-like buying experience at scale. We believe that in the near future, every online store will have AI employees managing sales and customer relationships — and Boom is building the platform that makes that possible.

Airports are among the most data-rich yet information-poor environments in transportation. Thousands of radio transmissions, aircraft movements, and sensor signals occur every second, but these systems remain siloed. Most airfields still rely on manual logs and calls to understand what’s happening on the ground, creating blind spots that slow coordination and limit safety oversight. The result is a growing safety and efficiency gap. The FAA records four to five runway incursions every day, and over 90% of air traffic control facilities are operating below target staffing levels. Despite $20 billion in modernization programs like NextGen and Safety Management Systems, airports still lack a unified data backbone that connects communication, movement, and intent across the airfield. Stratus provides the missing layer of intelligence for airport and airside operations — unifying data, communication, and movement into one shared, real-time picture. For airports, this means faster decisions, clearer accountability, and safer, more efficient operations. For aviation as a whole, it’s the foundation for connected, data-driven infrastructure that can finally keep pace with modern airside complexity.

AirCaps is bringing AI assistance to in-person conversations. Our AI-copilot provides live captions, translations, AI meeting notes and insights for in-person conversations in real-time. It's deployed as an app for lightweight AR glasses so you can see visual information overlaid onto your field of vision. You can think of it like a Zoom AI meeting assistant or Granola, but for in-person conversations, and with the ability to proactively help you in real-time, not just after the conversation. We’re building the capture and intelligence layer for the 216 billion daily conversations that happen face-to-face. We’ve already processed 16,500 hours of real-world conversations and counting. Today, AirCaps assists with 11% of our users’ in-person conversations. We did $93K in revenue in October and grew 6.5x from September while spending <$3K on marketing (and hit $70K in revenue in the first 2 weeks of November). Our power users average 6h+ daily usage and our day-30 retention is 91%. We previously went viral (75M+ views on TikTok, 150K+ followers) and have been featured by The New Yorker, WIRED, and Forbes. Madhav (CEO, Yale CS) has been building in audio and AR since age 13, starting with Google Glass apps and voice assistants for Raspberry Pis. He researched audio AI at the MIT Media Lab. Nirbhay (CTO, Cornell CS) built voice AI on smart glasses in high school and built conversational AI products as the first engineer at 2 YC startups.

Lapis is the most accurate AI search analytics platform. We help businesses understand the full buyer journey on AI assistants like ChatGPT and Perplexity, from discovery to purchase. We do this through a custom RL environment that's fine-tuned for AI assistant prompts.

Hyperspell (YC F25) gives AI agents memory. It connects to tools like Slack, Gmail, Notion, and Drive so agents can recall, reason, and remember across company knowledge. Developers use Hyperspell’s API to turn disconnected agents into AI coworkers that understand their work. Conor and Manu founded Hyperspell after building their own workplace agent and seeing how limited AI is without memory. They have over 15 years of experience in machine learning and have scaled API-based products to $30M ARR.

Semiotic builds AI design agents for companies. We’re starting with landing pages for fast‑growing startups. Existing AI website builders didn't make creating websites easier—they made creating ugly websites easier. So, companies choose between blue-purple gradient slop or firms with 3-month waitlists and $100k minimums. You don't have to choose anymore. Start a project: semiotic.so

We help construction teams eliminate rework, delays, and drawing confusion before construction starts.
We analyze PDF plans, detect clashes and inconsistencies across disciplines, and create draft RFIs so teams can resolve issues early. What we do: - Automatically detect clashes, missing information, and drawing mismatches - Align architectural, structural, and MEP plans so everything matches across sheets - Generate draft RFIs from issues we find - Replace long page-turn meetings with instant drawing analysis - Reduce rework and prevent schedule delays before boots hit the ground We integrate with:
Procore, Autodesk, email, and the tools teams already use Teams use us to: - Save hundreds of thousands to millions of dollars per project - Catch design and coordination issues early - Reduce RFIs and prevent field fixes - Keep projects on schedule with fewer surprises Contact us at founders@usearticulate.com

Compyle is a less autonomous coding agent that asks before it acts. We're like Cursor or Claude Code - but you won't need to rewrite your codebase every time you use it for more than a small task. We built Compyle because we were frustrated with the direction other coding agents are headed. They're all racing to be more autonomous, running for hours without human input. This doesn't work for real production code. When agents run autonomously, they make countless small decisions that pile up in your codebase. Engineers lose ownership - they don't know what was implemented, can't effectively debug or maintain it, and struggle to build on top of it. The code becomes a black box, and they're forced to throw it away. Compyle keeps engineers in the driver's seat.

Leadbay (YC F25) discovers and qualifies SMB leads that couldn't be found before. Small-to-enterprise teams use us to prospect the most low-data-low-signals industries like construction, hospitality, even startups. Co-founders are Milan Stankovic (CTO) - PhD to make AI work with limited data, and Ludovic Granger (CEO). Try us on leadbay.ai.

Cortex AI builds the world’s most diverse and large-scale real-world workplace robot & egocentric dataset — where the physical world becomes the next training and evaluation set for embodied AI. We power frontier labs developing robotics foundation models and general-purpose robots by providing the data they need: 1️⃣ Egocentric Data — real-workplace human video with hand/body pose, depth, and subtask labels. 2️⃣ Robot Data — trajectories collected from manipulators and humanoids in real industry settings. 3️⃣ Human-in-the-Loop Rollouts & Evals — real-world deployments with remote operators who recover robots when they fail, capturing data that feeds back into training and continuously improves models. Additionally, through the Cortex Marketplace, workplaces get paid to host data-collection and evaluation sessions, while labs access the in-the-wild data that truly matters. This draws on Lucas’s previous experience as co-founder of Carousell, a C2C marketplace that scaled to a $1B+ valuation.

Claybird makes studio-quality video ads in under 72 hours for brands like Coca-Cola, Walmart, Eight Sleep, Delve, and Mercor. Claybird’s AI agents use single-image NeRFs, fine-tuned image editing models, and eval layers for picture-perfect, consistent results at scale.

Ship reliable agents with confidence. AI agents often fail quietly, picking the wrong tools, looping, hallucinating, and more. The Context Company tells developers when and exactly where their agents fail so issues can be uncovered and fixed fast. It combines traditional observability across traces, tool calls, latency, and costs with the ability to uncover silent failures that standard monitoring misses. Get started with less than 10 lines of code.

Redapto builds self-improving systems for customer support teams that boost CSAT, NRR, and drive repeat purchases. Support quality plateau because agents and QA are static. Less than 5% of interactions are reviewed, missing SOP gaps, hallucinations, and outdated knowledge. Redapto audits 100% of interactions to detect issues, run custom evals, and auto-update SOPs, training, and knowledge bases so teams continuously improve. We’re applied researchers who’ve shipped this at scale. At Coinbase, our agent automations save $2.5M+ annually; at Adobe and Amazon, our post-training and compression work saved $1M+ and improved latency 20×.