VideoDB
VideoDB is real-time video infrastructure for AI agents. It ingests files, live RTSP streams, cameras, meetings and screen recordings, then indexes them into persistent, searchable memory that applications retrieve as playable clips through Python and JavaScript SDKs.
What is VideoDB?
VideoDB is a video backend built for software rather than for viewers. Published by Spext Labs Inc. under the VideoDB brand, it starts from the observation that the existing video stack, made of object storage, ffmpeg transcoding, HLS delivery, DRM and players, was designed so that people could watch, and that teams adding AI end up bolting transcription, vision models, embeddings, vector databases and custom orchestration onto it. VideoDB replaces that assembly with one system exposing six primitives: ingest, index, remember, retrieve, edit and stream.
The loop is the product. Any source enters, whether video files, YouTube URLs, live RTSP camera feeds, meetings, screen captures or datasets. Indexing runs over speech, scenes, people, objects, actions and events, each anchored in time. That understanding accumulates into persistent memory the application can query, instead of being recomputed on every request. Retrieval takes a plain-language question and returns the exact moments as playable clips. From there the platform acts: events and webhooks fire, agents are triggered, and new clips, captions, overlays or streams are generated by code.
The API surface is deliberately small, four concepts named Media, Indexes, Memory and Events, reached through Python and JavaScript SDKs, a REST API and an MCP server. The platform is model-agnostic: customers bring their own vision models and run them on every frame of a live feed. RTSP Connect, now generally available, handles live ingestion with rolling-window understanding and indexes that stay queryable while a stream is still running.
Four solution areas are documented: agentic perception, live camera intelligence, programmable media and training video data. Claimed figures include retrieval in roughly 120 ms across petabyte archives, 2.4 million hours indexed, 240 ms p95 query latency and 99.97% uptime over ninety days. Deployment comes in two tiers: a fully managed cloud with multi-region availability across the United States, the European Union and India, or a single-tenant deployment inside the customer's own AWS, GCP or Azure account, with on-premise on request and an optional edge GPU, using the same SDK and no code changes.
What it does
- Ingest video from files, YouTube URLs, live RTSP streams, cameras, meetings and screen recordings
- Index speech, scenes, people, objects, actions and events against their timestamps
- Search an entire archive in plain language and get playable clips back
- Keep the resulting understanding as persistent, scoped, queryable visual memory
- Trigger events, webhooks and agent actions the moment something matters on a live feed
- Generate clips, captions, overlays, dubbing and translations by code
- Serve every output as a native HLS stream URL, with no re-encoding
When to use VideoDB / When not to
A quick filter to help you decide if VideoDB is the right fit.
When to use VideoDB
- AI product teams building agents that need to see, remember and act on video
- Engineering teams running live camera operations, with thousands of RTSP feeds, custom computer vision models and sub-second alerts
- Media and OTT teams needing semantic search and programmable editing across large premium catalogues
- Machine learning teams mining raw footage for training samples, edge cases and labelled scenes
- Regulated buyers in healthcare, fintech, defence or telco who must keep video inside their own AWS, GCP or Azure account
When not to use VideoDB
- Anyone looking for a video generation tool: VideoDB explicitly does not create avatars or synthetic footage
- Non-technical users, since everything runs through Python or JavaScript code with no no-code interface
- Mobile-first users, as there is no iOS or Android application
- Creators wanting a consumer video editor, because editing is exposed as an API rather than a timeline
- Teams that require a documented opt-out from model training, which the published policies do not provide
How to use VideoDB
A typical end-to-end flow, from setup to results.
- Create an account on the VideoDB console and collect a free API key, with no credit card required
- Install the SDK with pip install videodb for Python, or npm install videodb for JavaScript
- For an agent runtime, bootstrap every primitive in one command with npx skills add video-db/skills
- Follow the quickstart in the documentation, which promises a first semantic search in about five minutes
- Add your first media: upload a file, point at a YouTube URL, or connect a live RTSP stream
- Build the indexes you need over spoken content, scenes, objects or events, or plug in your own model
- Query the index in natural language and collect the matching moments as playable clips
- Define events in plain language and wire the alerts to a webhook or a WebSocket listener
- Compose outputs by code, with inline edits, overlays, captions or dubbing, and serve them as HLS streams
- Move to your own AWS, GCP or Azure account when the workload requires it, keeping the same SDK calls
Pros & Cons
Pros
- One API and one SDK cover live streams and stored files with the same mental model
- Replaces an assembled stack of storage, transcoding, transcription, vision models, vector database and glue code with a single backend
- Model-agnostic by design, so customers bring their own computer vision or language models instead of being locked to the vendor's
- Deployment is portable, from managed cloud to the customer's own AWS, GCP or Azure account or on-premise, without code changes
- Compliance groundwork is already done, with SOC 2 Type II, ISO 27001, HIPAA and GDPR reports issued in November 2025
- Unit rates are published line by line, and a free tier with starting credits lets a team test without a sales call
- Open-source reference applications on GitHub give working starting points for screen capture, meetings and generative video
Cons
- Nothing is usable without writing code, and there is no mobile application
- Usage-based billing spread across a dozen separate line items makes the cost of a real workload hard to forecast
- Enterprise and bring-your-own-cloud pricing is quote-only, with no public figure
- No documented way to exclude customer data from the algorithm improvement the privacy policy permits
- Legal documents are inherited from the publisher's earlier Spext brand, so the Data Protection Officer answers on a third-party domain and the privacy text is written mainly around audio transcription
- The security page states that files are never stored while the privacy policy keeps video-related data for 180 days
- Compliance certificates are shared only on request under NDA, so nothing can be verified independently beforehand
Pricing & Plans
A permanent free plan is available at USD 0 per month, including USD 20 of starting credits and requiring no credit card. The lowest paid entry point is the Pro plan at USD 20 per month, which carries a rolling USD 20 monthly credit. Enterprise pricing is quoted on request. Beyond the plans, processing is billed at published unit rates, ranging from USD 0.01 per minute of transcription to USD 0.07 per gigabyte of video streaming.
- USD 20 of starting credits
- then pay as you go
- presented as best for testing the APIs
- USD 20 rolling monthly credit
- no rate limits
- unlimited credit top-ups
- auto-recharge
- priority support by email and Slack Connect
- tailored plan
- hybrid or on-premise deployment
- custom models and fine-tuning
- 99.9% SLA
- dedicated support team and expert consulting
Data, GDPR & hosting
A consolidated view of how VideoDB handles your data.
GDPR overview
GDPR compliance is claimed explicitly and backed by documents. The terms list the GDPR alongside HIPAA, ISO 27001 and SOC 2, and the trust centre offers a GDPR audit report issued in November 2025 by Intercert, shared under NDA on request. A Data Protection Addendum is published, built on the EU GDPR and incorporating the 2021/914 Standard Contractual Clauses, with a named list of subprocessors and their processing locations. The privacy policy, effective 30 October 2025, enumerates eight data subject rights and designates a Data Protection Officer, reachable at ashu@spext.co at the San Francisco address. Two gaps remain: personal data is transferred to and stored outside the European Economic Area, and no Article 27 EU representative is designated anywhere on the site.
Who owns the data?
Spext Labs Inc., operating under the VideoDB brand, is the data controller for its own site and account data, and acts as processor for the content customers upload under its published Data Protection Addendum. The terms state that uploaded content is confidential and proprietary to the customer. Access is restricted to authorised personnel, and employees are not authorised to view uploaded files during upload or transcription. Customers keep access to their files even after a subscription ends, since viewing, export and download remain available, and they can delete audio, video and transcripts at any time. Under the bring-your-own-cloud tier, originals never leave the customer's own storage.
Reuse rights
Customers keep the right to reuse their own material freely: nothing in the terms transfers ownership to VideoDB, and the outputs, whether clips, transcripts, indexes or streams, are derived from content the customer supplied. The vendor's own reuse is narrower but not nil. The privacy policy allows personal data contained in uploaded audio to be used to deliver the transcription service and to improve the company's algorithms, and permits aggregated, non-identifying statistics to be shared with affiliates and business partners. VideoDB states that it does not sell, rent or share information with third parties for promotional use, and that it does not disclose personal data except where compliance or the law requires it. Generative features route content to named subprocessors, including OpenAI and Google Cloud. No documented mechanism lets a customer exclude their data from that algorithm improvement.
Data retention & training
Hosting summary
VideoDB Cloud advertises multi-region availability across the United States, the European Union and India. The Data Protection Addendum names AWS as the data centre, with multiple availability zones and a multi-tenant environment in which customer data is logically isolated. Annex III lists every subprocessor with its processing location: Amazon Web Services, Google Cloud, OpenAI, AssemblyAI, SolarWinds, Linear and GitHub in the USA, Slack in the USA and the EU, and Cloudflare processing traffic globally at the data centre closest to the end user. The privacy policy is explicit that personal data is transferred to and stored outside the European Economic Area, and that it may move to offices and servers located in the United States. The publisher, Spext Labs Inc., is incorporated in Delaware and has offices in California and India. Customers who cannot accept that arrangement can take the bring-your-own-cloud tier, which deploys a single-tenant control plane inside their own AWS, GCP or Azure account, keeps originals in their own storage, uses private link and states no data egress; on-premise deployment is available on request.
Things to keep in mind
Risks and trade-offs to weigh before adopting VideoDB.
- Usage-based billing across a dozen line items, where recurring storage for media, spoken indexes and RTStream keeps charging every month and auto-recharge is switched on with the Pro plan
- The privacy policy permits personal data from uploaded audio to be used to improve the vendor's algorithms, with no documented opt-out
- The security page says files are never stored while the privacy policy retains video-related data for 180 days, a contradiction worth settling before uploading sensitive footage
- Personal data is transferred to and stored outside the European Economic Area, and no Article 27 EU representative is designated
- Generative features route customer content to named subprocessors including OpenAI and Google Cloud, which matters for confidential or regulated material
- Compliance certificates are available only on request under NDA, so no independent verification is possible before a commercial conversation
- Continuous camera and screen capture creates a standing surveillance capability: indexing people, actions and events at scale raises consent and proportionality questions the vendor leaves entirely to the customer
Setup & Integrations
Technical difficulty
Low for a developer, impossible for anyone else. Installation is a single package command, pip install videodb or npm install videodb, or one npx command to bootstrap an agent runtime. A free API key is issued from the console without a credit card or a sales call, and the vendor claims a first semantic search in about five minutes, with the API surface reduced to four concepts. Python or JavaScript fluency is the real prerequisite, as there is no no-code path. The bring-your-own-cloud deployment is a different matter and amounts to a genuine infrastructure project.
Deployment
Integrations
Behind VideoDB
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What exactly is VideoDB?
Does VideoDB generate video?
How is it different from calling a vision model API?
Can it handle live camera feeds?
How do I get started?
What does it cost?
Can I run it inside my own cloud?
Which certifications does VideoDB hold?
How long is data kept?
Who is behind VideoDB?
Should you pick VideoDB?
VideoDB is an infrastructure product and deserves to be judged as one. It does not generate video, it does not replace a vision model, and it offers nothing to someone unwilling to write code. What it offers instead is a single backend for the whole video loop: ingesting a file or a live RTSP feed, indexing speech, scenes and objects against time, keeping that understanding as queryable memory, returning exact moments as playable clips, and emitting new streams or events at the other end. Teams that would otherwise assemble object storage, ffmpeg, transcription, a vision model, a vector database and orchestration code are the intended audience, and the argument is coherent for them.
The commercial approach is unusually open for this category: unit rates are published line by line, a free tier carries starting credits, and the first paid step is USD 20 a month. The counterpart is that the cost of a real workload is genuinely hard to forecast across a dozen billed dimensions, and anything at enterprise scale returns to a quote.
The compliance groundwork is substantial: SOC 2 Type II, ISO 27001, HIPAA and GDPR reports dated November 2025, a published Data Protection Addendum naming every subprocessor and its processing location, and a bring-your-own-cloud tier that keeps originals inside the customer's own account. The weak points sit in the paperwork rather than the platform. The legal pages are inherited from the publisher's earlier Spext brand, the Data Protection Officer answers on a third-party domain, the security page and the privacy policy disagree over whether files are stored, and nothing documents an opt-out from the algorithm improvement the privacy policy permits. Serious buyers should settle those points in writing before committing sensitive footage.
- Choosing a selection results in a full page refresh.
- Opens in a new window.