# VideoDB > VideoDB is an applied lab for visual intelligence. It builds video infrastructure for agents and devices that see: an agent can upload or stream video, then search it, clip it and remember it. Four layers sit behind one API: primitives (streams, frames and in-house transcoding into a frame-addressable form), model-agnostic understanding (any model as an analyzer), indexes that make video searchable, and delivery (MCP tools, SDKs, WebSocket events, webhooks and a playable stream per result), run in VideoDB's cloud or the customer's. VideoDB exposes the whole site as Markdown alongside the HTML. Every page has a `.md` counterpart at the same path (`/agents` and `/agents.md`, `/blog/` and `/blog/.md`). Each HTML page declares this with `` in its ``. A request for any page with `Accept: text/markdown` is redirected to its `.md` counterpart. The whole site is also available concatenated at `/llms-full.txt`. ## Key facts - Use it from an agent: add the VideoDB plugin in Claude (https://videodb.io/claude) or ChatGPT (https://videodb.io/chatgpt), install the Claude Code plugin (`/plugin marketplace add video-db/skills`, then `/plugin install videodb@videodb-skills`), or add the remote MCP server `https://mcp.videodb.io/mcp` (Streamable HTTP) to Cursor or any MCP client and log in once. 70 tools: bring video in, understand it, search it, edit it, stream any result, watch live cameras and screens. - Free to start: $20 of free credits, no credit card, at https://console.videodb.io. - What people use it for: describing what is on screen, chapters and notes for lectures, finding moments, social clips, subtitles and translation, sports and coaching review, meeting follow-up, driving and security labelling, turning screen recordings into agent skills. - Visual memory: long-term memory is the index of everything seen; short-term memory is the frames a task needs, back in about half a second; real-time memory is the live stream. - Solutions built with teams, each with its own page: for teams with archival footage (sports, gaming, social media, advertising, meetings), agents that find any moment in the archive and cut, caption and publish it, plus social listening on what is on screen, retention next to what viewers saw, and scene context for ad breaks; computer-use agents built from recordings of how a team works, with every run recorded; for autonomous vehicle (AV) teams with cars and drones, every camera feed checked as it arrives, with the events that matter kept and routine miles moved to a cheaper bucket, plus scenario search over every drive and flight; for data providers, quality checks on every clip (technical quality, near-duplicates, variety) before the checked set is published as an MCP server (live example: https://datasets.bot); and for teams with live video (cameras, drones, screens), events described in a sentence, alerts with the clip behind them, and every feed searchable later. - The files stay in the customer's storage. VideoDB runs on top, on its own cloud paired with the customer's bucket, or entirely inside the customer's account. - Customers: teams that have paid for VideoDB include Capitol AI, Ezoic, Corvic AI, Virello, Clario, Muzology, BioRugged, Paradigm Life, Five Whys, Palette, IT STEP Academy, Turbine Workforce, Tailored AI and Seven Apps. Case studies: a Big Four accounting firm setting up its global capability centers with computer-use agents (https://videodb.io/blog/big-four-firm-gcc-agents.md), and 100,000 hours checked and cut into training samples in under a week for one of the largest video data providers (https://videodb.io/blog/video-data-provider-100000-hours.md). - Scale and quality: 100,000 hours of video turned into scene-level samples in under a week; 300+ TB of video processed; 1,000+ developers and agents use VideoDB every week. The retrieval engine leads on public benchmarks: see the technical report "Search over the Visual World" (arXiv 2608.08075) and the 100-hour visual moment retrieval benchmark (https://videodb.io/blog/a-faster-path-to-the-next-policy). - Install for code: `pip install videodb` (Python) · `npm install videodb` (Node.js) · `npx skills add video-db/skills` (coding agents). - Enterprise: SOC 2 Type II, GDPR with a DPA on request; managed cloud or the customer's own cloud, in any region. - Not video generation: VideoDB is video data infrastructure for agents (understanding, indexing, retrieval, realtime events, editing), not a video generator. ## Core - [Homepage](https://videodb.io/index.md): What VideoDB is: build agents with visual intelligence. How to try it in an agent, the teams using it every day and two case studies, visual memory, what teams build, the stack, open source and research. - [Add VideoDB to Claude, ChatGPT and any MCP client](https://videodb.io/agents.md): Add VideoDB to Claude, Claude Code, Cursor or any MCP client: install steps per client, the jobs people give the agent, the 70 tools, reference agents and questions. - [Agentic video primitives: streams, understanding, indexes, delivery](https://videodb.io/products.md): Agentic video primitives: primitives (streams, frames, transcoding), understanding, indexes and delivery, plus deployment in VideoDB's cloud or yours and the quickstart. - [Video hosting](https://videodb.io/video-hosting.md): Video hosting an agent can run: upload, list, transcode and stream from Claude, ChatGPT, any MCP client or the SDKs; what to ask and what comes back; what the agent can also do with a hosted video; the hosting rates; a Python example. - [Solutions: archives, computer use, AVs, datasets, live video](https://videodb.io/solutions.md): Where VideoDB works with teams directly: archival footage, computer use, autonomous vehicles, data providers and live video. - [Build agents for archival footage](https://videodb.io/archival-footage.md): Archival footage: agents that find any moment in years of games, streams, ads and meetings and turn it into a clip or a report; what sports, gaming, social media, advertising and meeting teams ask; a real session where Claude edited VideoDB's launch video; social listening, retention and ad-break context. - [Computer-use agents from how your team works](https://videodb.io/computer-use.md): Computer-use agents built from how a team works: the VideoDB Capture Agent records each step, about three weeks from recordings to a first agent, and every run kept on record. - [Autonomous vehicles: keep the seconds that matter](https://videodb.io/autonomous-vehicles.md): Autonomous vehicles (AVs), cars and drones: every camera feed checked as it arrives, the events that matter kept, routine miles moved to a cheaper bucket; scenario search over every drive and flight; scenario sets as manifests. - [Quality-check every clip, from every source](https://videodb.io/data-providers.md): Data providers: every clip from every source checked for video health, visual validity, camera stability, duplicates, variety and actions, then searchable by researchers' agents over MCP. datasets.bot is the live example. - [Live video: real-time alerts and search on every camera](https://videodb.io/live-video.md): Live video: cameras, drones and screens read as the video arrives; events described in a sentence, alerts over a webhook or a WebSocket with the reason and a clip, every feed searchable later; the flow as one Python file, a live demo on sample streams, notebooks per event, and a four-week pilot. - [Agent memory: long-term and short-term memory from live video](https://videodb.io/agent-memory.md): Agent memory from live video: each window of a feed is described, checked against events and indexed; long-term memory is that index, short-term memory is the frames a task needs, back as a stream in about half a second; querying by meaning, filter and count; a comparison with context windows, vector notes and clip pipelines; the recall in code; questions. - [Platform: a distributed data format for video](https://videodb.io/platform.md): The architecture (transcoding, frames, annotation, the distributed data format, retrieval, streaming, agent runtime), compliance, and the quickstart. - [Team: the people, investors and advisors](https://videodb.io/team.md): Team: the people building VideoDB, and the investors, advisors and institutions behind it. - [Pricing](https://videodb.io/pricing.md): Plans (Free, Pro, Enterprise), what Pro credits buy, the rate card for ingest, indexing, storage, editing, generation, delivery and tokens, and a cost calculator. - [Talk to the VideoDB team](https://videodb.io/contact.md): How to reach the team. ## Research & Blog - [Research & Blog index](https://videodb.io/blog.md): Papers, benchmark results, engineering notes, essays and apps from the VideoDB lab, published as the work happens. - Papers - [Search over the Visual World: persistent visual memory, layered indexes, and source-grounded evidence](https://videodb.io/blog/search-over-the-visual-world.md) (2026-08-08, Nagaonkar, Garg, Raj, Choithani, Trivedi): VideoDB technical report (arXiv 2608.08075). Analyzer-defined scenes, persistent visual memory, capability-declared indexes, source-grounded evidence, and a 9,834-query retrieval evaluation against a commercial video-native engine. - [Do thought streams matter? Evaluating reasoning in Gemini vision-language models for video scene understanding](https://arxiv.org/abs/2604.11177) (2026-04-13, paper, Sharma, Nagaonkar, Choithani, Trivedi): Benchmarks how internal reasoning traces affect video scene understanding in Gemini models, including where quality gains plateau and how tight budgets increase compression-step hallucination. - [Benchmarking vision-language models on optical character recognition in dynamic video environments](https://arxiv.org/abs/2502.06445) (2025-02-10, paper, Nagaonkar, Sharma, Choithani, Trivedi): Introduces an open-source benchmark for evaluating VLMs on OCR tasks in dynamic video environments across 1,477 manually annotated frames. - Benchmarks - [VideoDB leads a 100-hour visual moment retrieval benchmark against AWS Nova and Twelve Labs](https://videodb.io/blog/a-faster-path-to-the-next-policy.md) (2026-10-06, Sankalp Nagaonkar, Samuel Alexander): A head-to-head retrieval benchmark, and what it means for robotics teams choosing what their next policy should learn from. - [Can a VLM annotate a robot run? Results on a revised WGO-Bench](https://videodb.io/blog/wgo-bench-action-annotations.md) (2026-09-08, Samuel Alexander, Sankalp Nagaonkar): We re-annotated all 100 videos of WGO-Bench, human and robot, then measured how well a VLM pipeline segments and labels the actions in them. VideoDB reaches 28.93% semantic F1 at IoU 0.5, ahead of the public baseline at 19.92%, and the remaining errors show where the work goes next. - [What video retrieval benchmarks taught us about ground truth](https://videodb.io/blog/video-benchmark-ground-truth-ambiguity.md) (2026-07-24, Samuel Alexander): Manual review of MSRVTT, MSVD, VATEX, DiDeMo, and QVHighlights shows cases where the benchmark ground truth is too narrow, shifted, or ambiguous, so valid retrieved clips are scored as misses. - [A strong general-purpose VLM still fails on a chessboard](https://videodb.io/blog/claude-chessboard-spatial-reasoning.md) (2026-05-19, Sankalp Nagaonkar): We tested GPT-5.4 and Claude Opus 4.7 on reading 72 chess positions into FEN. Claude Opus 4.7 understood the boards and still lost to square-level errors, and the reason only showed up in the reasoning traces. - [Evaluating video VLMs on your own task](https://videodb.io/blog/how-to-evaluate-multimodal-vlms-for-your-video-use-case.md) (2026-05-15, Sankalp Nagaonkar): A practical workflow for evaluating video VLM setups with VideoDB and Langfuse, from task definition and dataset design to tracing, scoring, and deployment decisions. - [NFL game analysis: 80% fewer VLM hallucinations](https://videodb.io/blog/nfl-game-analysis-vlm-hallucinations.md) (2025-08-25, Sankalp Nagaonkar): Three approaches to event-dense NFL footage, measured. Play-by-play segmentation cut VLM hallucinations from 68.1% to 11.4% at up to 70% lower cost. - Customers - [A Big Four accounting firm sets up its global capability centers with agents](https://videodb.io/blog/big-four-firm-gcc-agents.md) (2026-10-30, VideoDB): One of the Big Four accounting firms uses VideoDB to turn recordings of expert work into SOPs and computer-use agents for its global capability centers, with every agent run on record. - [100,000 hours in under a week for one of the largest video data providers](https://videodb.io/blog/video-data-provider-100000-hours.md) (2026-10-30, VideoDB): VideoDB checked more than 100,000 hours of footage and cut it into scene-level training samples for one of the largest video data providers, in under a week, with the files left in the provider's storage. - Engineering - [How to evaluate a video AI platform: a PoC checklist](https://videodb.io/blog/evaluate-video-ai-platform-poc-checklist.md) (2026-10-26, Aditi Priya): Before a team commits to a video AI platform, it should try it on its own recordings. This checklist follows a robotics team running that trial, an AI proof of concept, from setting a pass/fail bar to making the final call. - [Understanding robotics datasets: inside LeRobot, RLDS and MCAP](https://videodb.io/blog/understanding-robotics-datasets.md) (2026-10-26, Om Gate): Open the files, follow one recorded step, and see how observations, actions and metadata fit together in LeRobot 3, RLDS and MCAP. - [Frame access that keeps research moving](https://videodb.io/blog/frame-access-that-keeps-research-moving.md) (2026-10-22, Samuel Alexander): How robotics researchers use VideoDB to select intervals from a run, sample frames for model analysis, and inspect the surrounding sequence. - [Twelve Labs alternatives, compared](https://videodb.io/blog/twelve-labs-alternatives.md) (2026-08-09, Ashutosh Trivedi): Comparing Twelve Labs alternatives for video AI in 2026: VideoDB, Mixpeek, Google, AWS, Azure, NVIDIA: organized by what you are actually building. - [Video RAG: how it works](https://videodb.io/blog/video-rag.md) (2026-08-09, Ashutosh Trivedi): What video RAG is, why it breaks text-RAG assumptions, the four architectures that work, and how to build one, including RAG over live streams. - [Focusd: a coach for your workday](https://videodb.io/blog/focusd.md) (2026-06-04, Sankalp Nagaonkar): How we built Focusd as a local-first desktop app that records work sessions, indexes screen activity with VideoDB, and turns raw events into useful productivity coaching. - [Dispatch #001: playable evidence, thinking tokens, agentic streams](https://videodb.io/blog/dispatch-001.md) (2026-05-08, Ashutosh Trivedi): The first issue of the VideoDB Dispatch: why agent runs need playable evidence, how much a video model should think, and builder updates from VideoDB. - [Reuse HTTP connections in Python services](https://videodb.io/blog/python-http-connection-pooling.md) (2026-04-20, Rohit Garg): A short field note on replacing repeated bare requests calls with a shared session so Python services can reuse HTTP connections. - [Caching CORS preflights: hundreds of milliseconds per call](https://videodb.io/blog/cors-preflight-cache.md) (2026-04-17, Om Gate): A practical note on preflight caching, custom auth headers, and why OPTIONS requests quietly dominate app latency on slower connections. - [A Postgres backup is not real until you restore it](https://videodb.io/blog/postgres-backup-restore-drill.md) (2026-01-02, Rohit Garg): A field note from a simple offline Postgres migration: why pg_dump was enough, where read-only mode fits, and why the restore drill matters more than the dump command. - [The 6 MB Lambda limit: compress before, not after](https://videodb.io/blog/lambda-compression-trap.md) (2025-09-17, Rohit Garg): Why API Gateway compression can make large responses look safe in testing, then still fail when Lambda enforces the raw payload limit first. - [Searching the slides in a conference recording](https://videodb.io/blog/conference-slide-extraction.md) (2025-08-25, Ashutosh Trivedi): Pull slide content out of any conference talk by combining spoken-word search with visual scene indexing. Full Python pipeline with working code. - [Smaller Python ML containers](https://videodb.io/blog/python-ml-container-size.md) (2025-08-25, Lalit Gupta): Slim bases, same-layer cleanup, CPU-only PyTorch, pinned wheels, and the unglamorous work of keeping deploys fast. - [VideoDB × TwelveLabs: search across your video library](https://videodb.io/blog/twelvelabs.md) (2025-08-13, Ashutosh Trivedi): TwelveLabs multimodal understanding plugs into VideoDB so any agent can find the exact moment it needs, not just the right file. - [VideoDB × LlamaIndex: video in your RAG pipeline](https://videodb.io/blog/llama-index.md) (2025-06-11, Ashutosh Trivedi): The official VideoDB connector for LlamaIndex lets you treat video as a first-class source in any retrieval-augmented pipeline. - Essays - [How a robot remembers](https://videodb.io/blog/how-a-robot-remembers.md) (2026-10-27, Ashutosh Trivedi): “Find my bottle and bring it to me.” A robot can only do this if it remembers what it has seen. This is a plain explanation of memory for robots and agents: the frames it keeps, the indexes that reach them, and how every task leaves the memory better than it found it. - [What is Physical AI? AI that acts in the real world](https://videodb.io/blog/what-is-physical-ai.md) (2026-10-14, Aditi Priya): Most AI today acts on a screen: it writes, codes and clicks. Physical AI gives AI hardware it can control, like a robot arm or a car's steering, and lets it act in the real world. This guide explains how it sees, decides, moves and learns. - [A query engine for robot video](https://videodb.io/blog/robot-runs.md) (2026-09-18, Ashutosh Trivedi): We built a query engine that sits over a lab's robot video, keeps every episode whole, and answers questions in plain English: find the failures, export them as a training manifest, and search the next model's runs the moment they land. - [DeepSeek’s visual primitives and the missing layer](https://videodb.io/blog/deepseek-visual-primitives-reference-gap.md) (2026-07-13, Sankalp Nagaonkar, Ashutosh Trivedi): The Reference Gap explains why seeing an object is insufficient when a multimodal model cannot preserve its identity, location, or path through a long reasoning trace. - [JEPA, from language models to world models](https://videodb.io/blog/jepa-from-language-models-to-world-models.md) (2026-07-07, Sankalp Nagaonkar, Ashutosh Trivedi): Why JEPA’s latent-prediction objective may shift AI systems from token prediction toward predictive world models for VLMs, VLAs, and embodied agents. - [Episodic memory for agents](https://videodb.io/blog/episodic-memory-for-agents.md) (2026-06-11, Ashutosh Trivedi): Humans remember experiences, not just facts. Your agent should too, with verifiable, playable evidence. - [Video was built for playback, not perception](https://videodb.io/blog/playback-vs-perception.md) (2026-06-09, Ashutosh Trivedi): 70 years of video infrastructure for human eyes, and why AI needs perception-first architecture. - [MP4 is the wrong primitive](https://videodb.io/blog/mp4-is-wrong-primitive.md) (2026-06-07, Ashutosh Trivedi): Video files were designed for playback. AI agents need indexes, not opaque blobs. - [Perception is the missing layer](https://videodb.io/blog/perception-is-the-missing-layer.md) (2026-06-05, Ashutosh Trivedi): LLMs have reasoning. RAG has retrieval. What's missing? The ability to perceive the world as it happens. - [Infrastructure that sees and edits](https://videodb.io/blog/infrastructure-that-sees-and-edits.md) (2026-06-03, Ashutosh Trivedi): Explore VideoDB's multimodal infrastructure that combines vision and editing for intelligent video processing. - [Why agents are blind today](https://videodb.io/blog/why-agents-are-blind.md) (2026-06-01, Ashutosh Trivedi): The gap between human perception and agent perception, and why it matters for the future of AI. - Agents - [How to add video to an app you built with Cursor](https://videodb.io/blog/add-video-to-cursor-app.md) (2026-11-18, Ashutosh Trivedi): Keep video files out of your repo. Host them with VideoDB from Cursor or Claude Code, then paste in an iframe or a small HLS player. The setup, the upload script and the React code are all here. - [Can Claude and ChatGPT Watch Videos? Yes, Here's How](https://videodb.io/blog/can-claude-chatgpt-watch-videos.md) (2026-11-18, Aditi Priya): On their own, Claude and ChatGPT see only part of a video. Connect VideoDB and they can search the whole thing, give you timestamps and cut clips of the moments you ask for. - [How to turn a video into a link with Claude or ChatGPT](https://videodb.io/blog/turn-video-into-link.md) (2026-11-18, Ashutosh Trivedi): Turn a video into a link you can share or embed, from Claude or ChatGPT with VideoDB added. Then get a unique link for any timestamp, or for a moment you describe, such as where pricing is discussed. - [Build your own video social-listening tool with VideoDB](https://videodb.io/blog/build-video-social-listening-tool.md) (2026-11-16, Aditi Priya): Most social listening tools read captions and comments. This guide shows how to build one that also hears and sees brand mentions inside videos: index a collection you supply, find every moment a brand is said or shown, and compare brands with a clip behind every number. - [Agentic video analysis: what works and what doesn’t](https://videodb.io/blog/agentic-video-analysis-what-works-and-what-doesnt.md) (2026-10-05, Aditi Priya): A video agent is a model that answers questions about video in steps, using tools. Some agents score well below a model answering on its own, and some score far above it. We look at what separates them: what the agent can search, whether it can look again, and whether it can check its answer against the footage. - [Why a search tool is not enough for an LLM working with video](https://videodb.io/blog/why-a-search-tool-is-not-enough-for-an-llm-working-with-video.md) (2026-10-05, Aditi Priya): A search tool lets an LLM find a moment in video. But when the LLM is asked how many, or only the ones that failed, it counts whatever the search returned. We show why that happens and what the LLM needs instead. - [Giving an agent a YouTube video it can search](https://videodb.io/blog/ai-agent-watch-youtube.md) (2026-08-09, Ashutosh Trivedi): Give your agent YouTube videos as searchable, timestamped context. Speech and visuals indexed, evidence clips compiled, in about 20 lines of Python. - [Claude edited our launch video with VideoDB skills](https://videodb.io/blog/claude-edited-our-launch-video.md) (2026-08-09, Ashutosh Trivedi): A case study in understanding-driven editing: the agent indexed the raw takes, picked the best take per narrative beat, composed and rendered the timeline, then re-indexed its own cut to audit it. - [Agentic Streams: an agent researches a topic and streams you the briefing](https://videodb.io/blog/agentic-streams.md) (2026-08-08, Ashutosh Trivedi): Open-source agents that research a topic on the live web, gather real clips, screenshots and charts, write a narration, assemble the video with VideoDB and hand back a stream you can play. Three agents included, with example outputs. - [Real-time visual perception for agents](https://videodb.io/blog/give-your-ai-agents-eyes.md) (2026-08-07, Ashutosh Trivedi): How AI agents get vision: ingest YouTube videos, live RTSP cameras, and screens; index continuously; search semantically; act on plain-language events. - [From an RTSP stream to events an agent can act on](https://videodb.io/blog/rtsp-ai-analysis.md) (2026-08-07, Ashutosh Trivedi): Turn any RTSP camera stream into AI-readable events: continuous understanding, plain-language alerts, and searchable history. No CV pipeline to build. - [TinyFish × VideoDB: the web your agent browses, as video it can search](https://videodb.io/blog/tinyfish.md) (2026-06-29, Ashutosh Trivedi): TinyFish opens the web for agents. VideoDB lets them see inside video. Together they cover both access and comprehension. - [Bloom: record your screen, search it later](https://videodb.io/blog/bloom.md) (2026-06-04, Lalit Gupta): How I built a local-first screen recorder on top of the VideoDB SDK. It automatically indexes your audio and screen, making everything searchable by the time you paste the share link. - [A camera feed for your OpenClaw agent](https://videodb.io/blog/openclaw-agent-camera.md) (2026-06-04, Ashutosh Trivedi): How we built a VideoDB-powered OpenClaw skill that records, indexes, searches, summarizes, and clips an agent's remote desktop without changing the agent itself. - [Deep Search: finding exact moments in video](https://videodb.io/blog/deepsearch.md) (2026-05-19, Sankalp Nagaonkar): How we built Deep Search as a retrieval loop for finding exact moments in video using planning, indexing, validation, recovery, and follow-up state. - [Pair Programmer: live screen and audio context for coding agents](https://videodb.io/blog/pair-programmer-live-context.md) (2026-05-19, Rohit Garg): A skill that gives Claude Code, Cursor and Codex live screen, mic and system-audio context, and how it is built on VideoDB Capture, RTStreams, local events and agent-side retrieval. - [How I built Call.md on VideoDB](https://videodb.io/blog/how-i-built-call-intelligence-with-videodb.md) (2026-05-08, Om Gate): A practical build story for a local-first call intelligence app that records calls, transcribes speakers, generates live nudges, and exports structured Markdown. - [Call.md, meeting copilot](https://videodb.io/apps/call-md.md) (app, built with VideoDB, Ashutosh Trivedi): Turn meetings into live agent loops. - [Bloom, AI-ready screen recorder](https://videodb.io/apps/bloom.md) (app, built with VideoDB, Ashutosh Trivedi): An agentic Loom alternative where recordings become inputs for AI. - [Focusd, productivity tracker](https://videodb.io/apps/focusd.md) (app, built with VideoDB, Sankalp Nagaonkar): Screen memory for understanding where your workday went. - [OpenClaw Monitoring](https://videodb.io/apps/openclaw-monitoring.md) (app, built with VideoDB, Ashutosh Trivedi): CCTV for computer-use agents. - [Cool QR Codes Studio](https://qr-generator-delta-lake.vercel.app/?utm_source=videodb_io&utm_medium=website&utm_campaign=videodb_side_projects) (app, side project, Ashish): A no-login QR code studio with scanability checks. - [Dictate](https://github.com/ash-ishh/dictate) (app, side project, Ashish): Local-first macOS dictation that pastes speech into any app. - Hackathons - [Unlock the Footage: July 2026 project showcase](https://videodb.io/blog/unlock-the-footage-july-2026.md) (2026-07-25, VideoDB): A global media intelligence hackathon where builders turned archived footage into evidence, memory, learning, and action. - [Agent Forge: June 2026 project showcase](https://videodb.io/blog/agent-forge-june-2026.md) (2026-06-13, VideoDB): A one-day, in-person build event where teams gave agents perception and action over real video and audio. - [Global Hackathon: May 2026 project showcase](https://videodb.io/blog/global-hackathon-may-2026.md) (2026-05-16, VideoDB): A 48-hour global build event exploring what agents can do when they can see, hear, and remember video and audio. ## Legal - [VideoDB Data Protection Addendum](https://videodb.io/legal/dpa.md): The standard Data Processing Agreement for customers processing personal data through VideoDB. - [Privacy Policy](https://videodb.io/legal/privacy.md): How VideoDB collects, uses, stores and protects personal data, and your rights as a data subject. - [Security](https://videodb.io/legal/security.md): How VideoDB protects customer data: encryption, access control, secure data centres and compliance. - [Terms & Conditions](https://videodb.io/legal/terms.md): Terms and conditions governing access to and use of VideoDB, the product of Spext Labs Inc. ## External - [Documentation](https://docs.videodb.io): SDKs, API reference, guides, with its own [llms.txt](https://docs.videodb.io/llms.txt). - [Console](https://console.videodb.io): get an API key. - [GitHub](https://github.com/video-db): SDKs, skills, Deep Search, the benchmark code, and the apps. - [PyPI](https://pypi.org/project/videodb/) · [npm](https://www.npmjs.com/package/videodb) - [Search over the Visual World on arXiv](https://arxiv.org/abs/2608.08075) - [Discord](https://discord.com/invite/py9P639jGz)