[
  {
    "id": 1910,
    "name": "Activeloop",
    "slug": "activeloop",
    "former_names": [
      "Snark.ai",
      "Snark AI"
    ],
    "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/c516ed5054847ecb1afb63f795f712b8d5c7f23d.png",
    "website": "https://activeloop.ai/",
    "all_locations": "San Francisco, CA, USA; Mountain View, CA, USA",
    "long_description": "We provide a simple API for creating, storing, versioning, and collaborating on multi-modal AI datasets of any size. With Activeloop's open-core stack, you can rapidly transform and stream data while training models at scale. Deep Lake powers foundational model training by acting as a vector database with significant benefits, such as (1) the ability to use multi-modal datasets to fine-tune your own LLM models, (2) storing both the embeddings and the original data with automatic version control, so no embedding re-computation is needed (3) truly serverless service with no vendor lock-in. How cool is that?\r\n\r\nGitHub loves us - we're one of the fastest-growing libraries there, and we're used by little-known companies like Google, Waymo, and Intel. No big deal. \r\n\r\nOur founding team hails from places like Princeton, Stanford, Google, and Tesla, and we're backed by Y Combinator & other Silicon Valley heavyweights. \r\n\r\nActiveloop is hiring, and we want you! Check out our open roles on our YC page and join the fun.\r\n\r\n10-min demo: https://activeloop.wistia.com/medias/aibvo0dst2\r\nWhitepaper: https://www.deeplake.ai/whitepaper",
    "one_liner": "Database for AI",
    "team_size": 15,
    "industry": "B2B",
    "subindustry": "B2B -> Infrastructure",
    "launched_at": 1524690140,
    "tags": [
      "Computational Storage",
      "Deep Learning",
      "Generative AI",
      "Computer Vision",
      "Open Source"
    ],
    "tags_highlighted": [],
    "top_company": false,
    "isHiring": true,
    "nonprofit": false,
    "batch": "Summer 2018",
    "status": "Active",
    "industries": [
      "B2B",
      "Infrastructure"
    ],
    "regions": [
      "United States of America",
      "America / Canada",
      "Remote",
      "Partly Remote"
    ],
    "stage": "Growth",
    "app_video_public": false,
    "demo_day_video_public": false,
    "app_answers": null,
    "question_answers": null,
    "url": "https://www.ycombinator.com/companies/activeloop",
    "api": "https://devasheeshg.github.io/yc-api/batches/summer-2018/activeloop.json",
    "year_founded": 2018,
    "city": "San Francisco",
    "country": "US",
    "linkedin_url": "https://www.linkedin.com/company/activeloop",
    "x_url": "https://twitter.com/activeloopai",
    "fb_url": "https://www.facebook.com/activeloopai",
    "cb_url": "https://www.crunchbase.com/organization/snark-ai",
    "github_url": "https://github.com/activeloopai",
    "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/c516ed5054847ecb1afb63f795f712b8d5c7f23d.png",
    "app_video_url": null,
    "dday_video_url": null,
    "company_photos": [],
    "primary_partner": {
      "name": "Diana Hu",
      "url": "https://www.ycombinator.com/people/diana-hu"
    },
    "founders": [
      {
        "user_id": 78090,
        "full_name": "Davit Buniatyan",
        "title": "Founder",
        "founder_bio": "Founding CEO Activeloop, PhD on leave from Princeton, AI/ML, Data and Infra, Y Combinator S18, UCL 16’ Working on Data 2.0",
        "is_active": true,
        "linkedin_url": "https://www.linkedin.com/in/davidbuniatyan",
        "x_url": "https://twitter.com/DBuniatyan",
        "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/49efe4795d8a38cc955a0c873f7de567ef057a52.jpg"
      }
    ],
    "jobs": [
      {
        "id": 79215,
        "title": "Member of Technical Staff",
        "url": "https://www.ycombinator.com/companies/activeloop/jobs/zn1plZe-member-of-technical-staff",
        "location": "Mountain View",
        "type": "Full-time",
        "role": "Engineering",
        "role_type": "Full stack",
        "salary_range": "$150K - $250K",
        "equity_range": "0.50% - 1.50%",
        "experience": "Any (new grads ok)",
        "visa": "US citizen/visa only",
        "skills": [
          "Amazon Web Services (AWS)",
          "Google Cloud",
          "Kubernetes",
          "MongoDB",
          "PostgreSQL",
          "Python",
          "Redis",
          "SQL",
          "Docker",
          "Microservices"
        ]
      }
    ],
    "news": [
      {
        "title": "Activeloop-L0: Agentic Reasoning on Your Multimodal Data",
        "url": "https://www.activeloop.ai/resources/activeloop-l-0-agentic-reasoning-on-your-multimodal-data/",
        "date": "May 04, 2025"
      },
      {
        "title": "Introducing Deep Research for Your Multi-Modal Data",
        "url": "https://www.activeloop.ai/resources/introducing-deep-research-for-your-multi-modal-data/",
        "date": "Feb 10, 2025"
      },
      {
        "title": "Releasing Deep Lake 4.0: The Fastest Multi-Modal AI Search on Data Lakes",
        "url": "https://www.activeloop.ai/resources/deep-lake-4-0-the-fastest-multi-modal-ai-search-on-data-lakes/",
        "date": "Oct 17, 2024"
      },
      {
        "title": "Activeloop Named 2024 Gartner Cool Vendor in Data Management",
        "url": "https://www.activeloop.ai/resources/gartner-cool-vendor/",
        "date": "May 31, 2024"
      },
      {
        "title": "Activeloop raises $11M to grow its specialized tensor database for AI training and inference - SiliconANGLE",
        "url": "https://siliconangle.com/2024/03/26/activeloop-raises-11m-grow-specialized-database-ai-training-inference/",
        "date": "Mar 26, 2024"
      }
    ],
    "launches": [
      {
        "id": 102722,
        "title": "Hivemind: Continual Learning for Coding Agents",
        "tagline": "Capture every coding agent interaction across your team, train it into reusable skills, and propagate those skills to every agent in your org. On your cloud storage.",
        "body": "Hey 👋 Davit here, founder of Activeloop.\n\nToday we are introducing something we believe will reshape how engineering teams work with coding agents.\n\n**Hivemind is a continual learning layer for coding agents. It works across Claude Code, Codex, Cursor, OpenClaw, Hermes, and pi, and it stores all your data on your cloud.**\n\n<https://youtu.be/rja1DfjUdww>\n\n### The Problem: Coding Agents Don't Learn\n\nEvery engineering team building with AI agents already knows the issues.\n\n* Your senior engineer's agent debugs a tricky bug Monday. Your junior engineer hits the same bug Tuesday and starts from zero.\n* Every session begins with re-explaining your codebase and your conventions.\n* Architectural decisions made in code review evaporate the moment the session ends.\n* Your team is solving the same problems over and over, paying for the same compute every time.\n\nCoding agents do not communicate with each other out of the box. Hivemind closes the gap. An agent figures something out, every agent inherits it. Your worst run starts from your best.\n\nThe tools tackling this treat it as a memory problem. They give each agent a personal notepad. Siloed. Invisible to the rest of the team. And most of them do it by shipping your code and context to their servers.\n\nBut memory without learning doesn't make agents better.\n\n**We decided to go beyond memory.**\n\n![uploaded image](/media/?type=post&id=102722&key=user_uploads/78090/51dc4451-7936-4e37-a0ad-e8139a646430)\n\n### Introducing Hivemind: Continual Learning for Coding Agents\n\nHivemind runs on a simple chain: **capture, codify, optimize, propagate.**\n\n**1. Trace capture.** Every agent interaction (prompts, tool calls, file reads, reasoning chains, outputs) is captured automatically as a structured trace.\n\n**2. Skill codification.** Repeated patterns across traces get codified into reusable skills. Mix of automatic detection and LLM-assisted extraction, with workspace-level scoping so skills don't leak between teams.\n\n**3. Skill optimization.** This is new today. Skills get trained and optimized. We've implemented SkillOpt (a text-space optimizer out of Microsoft, Shanghai Jiao Tong, and Fudan) directly into Hivemind. It improves each skill the way you'd tune a model, keeping only the edits that prove out on a held-out test.\n\nThat approach lifts agent accuracy by +19.1 points inside Claude Code and +24.8 inside Codex, and wins or ties on all 52 setups tested. It runs offline, so it adds zero cost at inference. Your skills get sharper over time instead of bloating.\n\n**4. Skill propagation.** Optimized skills flow into every agent's context at inference time, across every Hivemind-connected agent in your workspace.\n\nWorks across:\n\n* Claude Code\n* Codex\n* Cursor\n* OpenClaw\n* Hermes\n* pi\n\n![uploaded image](/media/?type=post&id=102722&key=user_uploads/78090/e1bddec6-5c2c-4409-aa2a-dd28fd8b2405)\n\n### On Your Cloud\n\nThis is the part our customers care about most. Your traces, your skills, your cloud storage. The other tools in this space treat your codebase as their training data. We don't. Hivemind store data on your cloud storage bucket, which is the only way a continual learning layer should ever touch your source.\n\n### **Codebase knowledge graph:**\n\n* Hivemind builds a graph of your codebase\n* Agents reason over how files, functions, patterns, and prior fixes connect\n* Structure beats keyword search: agents retrieve relevant context instead of grepping blindly\n\n![uploaded image](/media/?type=post&id=102722&key=user_uploads/78090/414a60d3-f2b1-4da0-8bec-236e44f16042)\n\n### Example\n\n> Your senior engineer's agent figures out a tricky migration pattern in your payment service on Monday.\n>\n> Tuesday, your junior engineer's agent hits the same migration shape in a related service. The skill has already crystallized, and SkillOpt has already sharpened it against the cases where the first version fell down. The agent executes it in one shot.\n\nRepeat work stops being repeat work.\n\n### The Results: 25% Lower Cost, 41% Fewer Tokens\n\nWe benchmarked Hivemind against memory tools using LoCoMo, the standard evaluation for long-horizon agent memory.\n\nHivemind matches Mem0 on accuracy.\n\n* **25% lower cost per 100 QA**\n* **41% fewer output tokens**\n* **31% fewer agent turns**\n\nSingle-agent benchmarks understate the real story. The compounding effect of org-wide skill propagation isn't captured in these numbers. In production, the gap widens every week as your trace library grows.\n\n![uploaded image](/media/?type=post&id=102722&key=user_uploads/78090/3df657fa-e32b-4329-a667-d055d39d93be)\n\nFull methodology: [deeplake.ai/hivemind](http://deeplake.ai/hivemind)\n\n### Why This Matters\n\nMost developer tools deliver linear value. Hivemind compounds.\n\n* **Week 1:** your agents stop repeating mistakes.\n* **Month 1:** they're learning from each other.\n* **Quarter 1:** your whole engineering org is operating with capability that survives team changes, onboarding cycles, and turnover.\n\nYour team's hard-won knowledge accumulates.\n\n![uploaded image](/media/?type=post&id=102722&key=user_uploads/78090/6dfbd341-418b-4bd6-bd63-0d191625378c)\n\n### What's Next: From Skills to Custom Models\n\nBecause we store trajectories in Deeplake's tensor format, traces are ready as PyTorch datasets. A handful of advanced customers are already fine-tuning their own open-source models on the trajectories their Claude and Codex agents generated last week.\n\nSame data layer. Two paths to continual learning.\n\nFull tutorial and public examples shipping in the coming weeks.\n\n### Get Started\n\nTry Hivemind: [deeplake.ai/hivemind](http://deeplake.ai/hivemind)\n\nGitHub: [github.com/activeloopai/hivemind](http://github.com/activeloopai/hivemind)\n\nSkillOpt paper: [arxiv.org/pdf/2605.23904](http://arxiv.org/pdf/2605.23904)\n\nThree commands and your team's agents share one brain by end of day.",
        "url": "https://www.ycombinator.com/launches/Qio-hivemind-continual-learning-for-coding-agents",
        "votes": 9,
        "created_at": "2026-06-08T23:07:52.598Z"
      },
      {
        "id": 95919,
        "title": "Activeloop Scientific Discover: The AI Analyst for Scientific Research",
        "tagline": "Search, reason, and discover across 450M+ pages and 175TB of scientific research papers. Not just text, but charts, molecules, formulas, and tables.",
        "body": "Hey 👋 Davit here, founder of Activeloop.\n\nToday we are introducing something we believe will fundamentally reshape how scientific discovery happens.\n\n**Activeloop’s Scientific Discover, an intelligence agent built on one of the largest datasets of indexed scientific research. Here are the details:**\n\nhttps://www.youtube.com/watch?v=x8Lv5-C9ntw\n\n## The Problem: Scientific Data Is Broken\n\nEvery scientist already knows the pain.\n\n* Critical insights are buried across PDFs, images, tables, and supplementary files.\n* Key figures live inside screenshots, not structured databases.\n* Twenty years of research is trapped in incompatible formats and outdated systems.\n* Every new project begins with weeks of manual data wrangling.\n\nThe White House recently launched the Genesis Mission, recognizing that fragmented scientific data is one of the biggest blockers to breakthroughs in drug discovery, materials science, climate modeling, and more.\n\nThe Genesis Mission aims to build an integrated AI platform that harnesses Federal scientific datasets. The goal is to train scientific foundation models and create AI agents that test hypotheses, automate research workflows, and accelerate scientific breakthroughs.\n\nThe country is trying to unify scientific data for AI.\n\n**We decided to show what that future looks like.**\n\n## Introducing L1: The Scientific Data Agent\n\nTo demonstrate what becomes possible when scientific data is indexed natively for AI, we built **Activeloop L1**, a multimodal scientific research agent running on top of:\n\n* 25M open access papers\n* 450M+ plus pages\n* 175TB of fully visual indexed scientific data\n\nUnlike traditional NLP or paper search, **L1 sees science**.\n\nIt understands:\n\n* charts\n* molecules\n* protein structures\n* equations\n* experimental tables\n* clinical graphs\n* and the text that surrounds them\n\nThis allows L1 to answer questions no text only model can touch.\n\n## The Results: A New State of the Art **48% SOTA on Humanity’s Last Exam with Tools**\n\n![uploaded image](/media/?type=post&id=95919&key=user_uploads/78090/453774e7-7cc6-47de-a344-76a9f7ca2460)\n\nL1 (Gemini 3 Pro) outperforms top models including Grok Heavy with tools, and GPT 5 Pro with tools.\n\n### **Multimodal scientific reasoning**\n\nL1 synthesizes biochemical structures with clinical findings to answer end-to end research questions.\n\n### Example Query\n\n> Which compounds show novel synergy with metformin for treating type 2 diabetes?\n\nL1 scans molecular structures, clinical outcomes, and experimental tables at the same time and produces grounded, citation linked answers.\n\n## Why This Matters\n\nMultimodal scientific agents unlock capabilities that were not possible before.\n\nThis extends far beyond drug discovery. It applies to materials science, climate research, algorithmic innovation, and more.\n\nThe applications are enormous.\n\n**Try L1 Today with Our OpenAI Compatible API**\n\n![uploaded image](/media/?type=post&id=95919&key=user_uploads/78090/f45683d6-c74b-4994-88e7-d54bda796ba0)\n\n## Get Started\n\nTry the Science Agent: https://chat.activeloop.ai/science\n\nDocs and API: https://docs.activeloop.ai\n\nBuild with the OpenAI compatible API today.",
        "url": "https://www.ycombinator.com/launches/Ox5-activeloop-scientific-discover-the-ai-analyst-for-scientific-research",
        "votes": 29,
        "created_at": "2025-12-02T15:44:42.819Z"
      },
      {
        "id": 94768,
        "title": "Activeloop - Unlock AI Data Analysis for GTM",
        "tagline": "Less Time in CRM, More Sales. Get instant insights across your entire stack from CRMs/ERPs to sales decks and product analytics. ",
        "body": "Hey 👋! Founder of Activeloop here. I am excited to introduce Activeloop AI Data Analysis for GTM operations. It lets you drill down from an executive dashboard to a single data point with just-in-time intelligence.\n\nWe didn’t built yet another dashboarding tool. It’s a solution to the single biggest bottleneck holding back your commercial strategy. The endless cycle of manual data preparation, integration, and reconciliation.\n\nhttps://www.youtube.com/watch?v=4Ua3Llox2HE\n\n### The Problem: You’re Drowning in Data Chaos\n\nThe story is always the same:\n\n* **Analysts as Data Janitors:** Your most valuable analysts are buried in manual work, stitching together data from multiple CRMs, reconciling Salesforce with SAP, and cleaning up messy, inconsistent fields.\n* **Reporting Backlogs:** A simple request for a new report kicks off a multi-week process. By the time it’s ready, the moment has passed and leadership doesn't trust the numbers anyway.\n* **Sales Reps Suffer:** Reps waste 4-6 hours every week just hunting for information and doing administrative work instead of selling.\n\nYou're tasked with driving strategy, but you're stuck managing data chaos.\n\n![uploaded image](/media/?type=post&id=94768&key=user_uploads/78090/e282c1b1-d361-4f5a-a4ac-3a721b21d96d)\n\n### Our Solution: The AI Analyst that Vibes Intelligence\n\nActiveloop is an AI Data Analyst for your operations team. We automate the painful, time-consuming data work that currently consumes 70%+ of your team's time.\n\nInstead of fighting with brittle integrations, we connect your fragmented sources such as CRMs, ERPs, product analytics, databases, data warehouses, even unstructured contracts and call transcripts, into a unified, queryable resource.\n\n![uploaded image](/media/?type=post&id=94768&key=user_uploads/78090/342debf4-c4fb-4a02-8480-ea5b128e26c9)\n\n### From Weeks to Minutes: Deliver “Just-in-Time” Intelligence\n\nActiveloop eliminates the wait, empowering your team with true self-service analytics.\n\n* **For Leadership:** In a forecast meeting, infinitely drill down into the data with simple questions like, \"Why did this region's deal cycle lengthen last quarter?\" Get immediate answers, without ever having to say, \"We'll get back to you.\"\n* **For Operations:** Free your analysts from being \"report builders\" and empower them to be strategists. They can finally focus on high-impact projects, confident that the underlying data is unified and trusted.\n\n![uploaded image](/media/?type=post&id=94768&key=user_uploads/78090/f5a7e592-3655-494e-a620-f92e99dc74e9)\n\n### We Tackle Your Messy Data Head-On\n\nYou’re probably thinking: \"How can AI possibly navigate our messy data? We have 50 different fields for the same metric.\"\n\nYou're right. That's the biggest blocker for any data initiative. That’s why our platform includes an **AI-assisted data mapping and taxonomy management layer**. It helps you harmonize your complex environment to build a clean data foundation that generates insights everyone trusts.\n\n![uploaded image](/media/?type=post&id=94768&key=user_uploads/78090/22c38e19-0d94-49cb-9485-e5de80ce8e3d)\n\n### Let's Solve Data Chaos Together\n\nIf you’re tired of seeing your team buried in manual data work, we want to talk to you.\n\n**Book a Call:** https://activeloop.ai/sales/\n\n\\\n**Get Started Now:** https://chat.activeloop.ai",
        "url": "https://www.ycombinator.com/launches/OeW-activeloop-unlock-ai-data-analysis-for-gtm",
        "votes": 107,
        "created_at": "2025-10-20T15:41:53.206Z"
      },
      {
        "id": 90294,
        "title": "Activeloop L0: State-of-the-Art RAG Accuracy on Your Data",
        "tagline": "Turn PDFs, images & tables into instant, cited answers. Agentic RAG that just works.",
        "body": "Hey! Davit here form Activeloop. While working on [Deep Lake](https://github.com/activeloopai/deeplake), I have seen many RAG systems collapse when exposed to production-scale corporate data. They often rely on predefined loops, custom logic and rigid agent scaffolds. Activeloop-L0 provides your agent with highly precise and answers grounded on your multimodal data.\n\nhttps://youtu.be/pnU9yd4KWy0\n\n**Why can’t we reliably analyze corporate documents?**\n\n* **Architectural hurdles**: messy data integrations, unexpected infra costs, and reliability/safety constraints.\n* **Commodity RAG** lacks depth for multimodal enterprise data (documents, images, audio).\n* **Infrastructure burden**: parsing, chunking, embeddings, indexing, vector DBs, and agent loops slow teams.\n\n**But wait, is RAG still relevant despite large context models?**\n\nLet’s consider four extensive NASA documents \\[[1](https://www.nasa.gov/wp-content/uploads/2022/03/sls-reference-guide-2022-v2-508-0.pdf), [2](https://www.nasa.gov/wp-content/uploads/2023/02/orion-reference-guide-111022.pdf), [3](https://www.lpi.usra.edu/lunar/artemis/Artemis-I-Reference-Guide_NP-2022-03-3045-HQ.pdf), [4](https://www.ulalaunch.com/docs/default-source/rockets/2023_vulcan_user_guide.pdf)\\], each between 80 to 100 pages, containing visual descriptions, and pose a highly complex question.\n\n![uploaded image](/media/?type=post&id=90294&key=user_uploads/78090/c7a09db6-f7f7-46a4-a347-5b52154e6fc1)\n\nChatGPT with o3, despite having full PDFs in context, failed after 11 minutes of reasoning. Now, imagine you have thousands of corporate documents that can’t be contained in a context. In contrast, Activeloop-L0 provided the correct answer in 4 minutes and can scale to a million documents.\n\n**What is Activeloop-L0?**\n\n**Activeloop-L0** is a compound AI system that ingests your unstructured data and returns grounded answers. Behind the scenes, [Deep Lake](https://github.com/activeloopai/deeplake) indexes neural representations at scale, then fuses “thinking tokens” with high-precision retrieval for fast multi-hop reasoning.\n\nIt is available on [**chat.activeloop.ai**](http://chat.activeloop.ai) now.\n\n![uploaded image](/media/?type=post&id=90294&key=user_uploads/78090/64161f83-f7ba-46eb-9445-5688f55506ab)\n\n**How is it different compared to a traditional RAG?**\n\n* **Multimodal:** Built-in support for images, PDFs, audio, and spreadsheets.\n* **Integrated Reasoning & Retrieval:** Eliminates the need for loops.\n* **Deep Indexing:** Cost-effective multi-layer indexing for richer context early on.\n* **Simple:** Focus on innovation, not maintaining infrastructure.\n* **Grounded and Accurate:** Clear citations for trustworthy insights.\n\n**How accurate is Activeloop-L0?**\n\nActiveloop-L0 achieves overall 85.6% state-of-the-art accuracy on 1,142 multimodal questions (292 PDFs, 5.5K pages). It outperforms text only RAG by +20%, visual RAG by +10%, and Alibaba’s ViDoRAG by +6% on their own ViDoSeek benchmark.\n\n![uploaded image](/media/?type=post&id=90294&key=user_uploads/78090/8275782e-ffde-4351-bdc8-0d1c40918169)\n\n**Is there an OpenAI-compliant API?**\n\nYes, Activeloop-L0 is available with an OpenAI-compliant API. You can easily plug into your agents for providing high relevant context. You can get started here. https://docs.activeloop.ai/setup/quickstart\n\n![uploaded image](/media/?type=post&id=90294&key=user_uploads/78090/996c0193-4dab-49eb-ab85-6e78828b19ec)\n\n**Ready to Deploy on Your Data?**\n\nActiveloop is trusted by F500 including the likes of [**Bayer**](https://www.activeloop.ai/usecase/bayer/), [**Flagship Pioneering**](https://www.activeloop.ai/usecase/flagshippioneering/), [Matterport](https://www.activeloop.ai/usecase/matterport/) (W12 acquired by CoStar).\n\n* **Your Cloud**: Deploy on your cloud, ensuring data never leaves your infrastructure.\n* **Your Models**: Integrate your LLMs.\n* **Your Security**: SOC2 compliance, fine-grained access control, and SSO.\n\n[**Book a call**](https://www.activeloop.ai/contact/) to discuss enterprise deployment.",
        "url": "https://www.ycombinator.com/launches/NUM-activeloop-l0-state-of-the-art-rag-accuracy-on-your-data",
        "votes": 34,
        "created_at": "2025-05-12T16:26:50.543Z"
      }
    ]
  }
]