Vivu product reference for agents

Last updated: 2026-10-12.

Vivu builds video intelligence for teams, AI workflows, edge devices, and robotics. Its hosted product makes recorded video searchable by what is shown, said, and written on screen. Its remote Model Context Protocol (MCP) server brings that search into connected assistants. Vivu also provides an optimized model for real-time video detection and indexing on edge devices, and helps robotics teams improve vision-language model (VLM) object recognition and relationship understanding.

This reference covers the company, product capabilities, use cases, integrations, access, pricing, operating behavior, and public resources. The three offerings below have different workflows and access models.

Product overview

The published remote MCP workflow is for hosted video projects and search. On-device deployments and robotics VLM optimization are separate offerings; their integration details are scoped with the team. The hosted subscription table does not establish pricing or entitlement for those engagements.

Main website: Vivu. Product details: Platform. Hosted application: Vivu web app.

Company

Vivu is a Seattle-based team building a connection between video and the software, teams, and agents that need to use it. Its work began with a recurring problem: companies had substantial libraries of footage, webinars, events, customer stories, and calls, but could not reliably retrieve the moments inside them.

The team describes more than ten years of video AI experience, including work at Microsoft, Unity, and earlier startups. Its About page attributes reach of more than ten million users and processing across more than one hundred million hours of video to technology the team built previously. Those figures describe prior experience, not the current user count or processing volume of Vivu.

Sources and company resources: About, Press and mentions, LinkedIn, YouTube, and Instagram.

The Press page links to coverage and community mentions. Historical coverage describes the product at its publication date; current product pages and documentation describe the current offering.

Hosted video intelligence

Video Context Graph

Vivu calls its structured video context layer the Video Context Graph, or VCG. It indexes video content rather than relying only on filenames, folder organization, manual tags, or upload dates.

The public Platform description includes visual content, speech, speakers, text on screen, and when those signals occur. The context layer connects moments and their surrounding information so teams and agents can query footage by meaning.

The hosted workflow indexes a recording when it is ingested and reuses the index for later searches. Search still consumes search credits. Reusing the index does not imply that later queries are free or that search spans every project in an account.

The VCG description is a product-level explanation. The MCP reference below describes the concrete publicly documented workflow and its project scope. See Platform.

What search finds

A query can describe an object, action, scene, spoken topic, visible text, or a combination of signals. Examples include a customer explaining a purchasing decision, a product shown in a particular shot, an error screen in a support clip, or an event in a recorded robot run.

Search returns candidate video moments as time ranges associated with source recordings. Results can miss relevant moments or include unsuitable ones. Opening and checking the original footage is part of selecting reliable evidence.

Vivu's hosted search locates existing footage. Editing a sequence, making a finished video, or generating new video belongs to other tools. Preparing an original segment for download is distinct from editing or generating a new video.

Projects and workspaces

Videos belong to projects. Each search runs inside one project, and the documented connector does not provide cross-project search. Keeping different clients or datasets in separate projects provides a clear search boundary.

The web app and connector use the same account, projects, and allowance. A project created in the web app can be searched through a connected assistant using the same account.

The pricing page describes personal and organization workspaces. Organization projects are visible to members of that organization workspace, while private projects are not. Teams interested in organization workspaces can discuss their needs with Vivu. There are currently two named self-service plans, Free and Premium; a separate named Enterprise subscription is not listed.

See Documentation and Pricing.

Upload and indexing

Hosted search uses video uploaded through a browser into a Vivu project. The upload page can be reached directly from the application or opened through an assistant using the MCP connector.

Upload and indexing are separate stages. Indexing runs asynchronously in Vivu's cloud, and a file is searchable only after indexing finishes. A successful upload alone does not mean the video is ready for search.

The secure upload sign-in link must be opened within three minutes. That window is for opening the link, not a maximum video duration. The browser page needs to remain open until the file transfer finishes.

The public documentation describes support for common video formats. For hosted search of robot recordings, camera streams contained in ROS bag or MCAP logs need to be exported to video files before upload.

Sources: Connector reference, MCP upload workflow, and Robotics.

Search modes and results

A search needs a project and a description of the desired moment. Mode is optional. When Precise search is used, completion is checked on the same search job before presenting a finished result; starting repeated searches is not necessary to obtain the existing job's outcome.

The result page groups returned moments for preview. To download a segment, select a result and use the existing Export original clip action. Result-page links expire after four hours; a fresh link is needed after expiry.

Sources: Search reference, Opening a moment, and Credits.

On-device video intelligence

Vivu's optimized on-device model performs real-time video detection and indexing locally on edge devices. Example device categories include phones, Raspberry Pi systems, and robots.

On-device means inference runs on the target device itself. The core capability described here is video detection and local indexing as video is processed. Specific deployments are scoped to the target hardware and workload.

This capability is broader than a phone application. A mobile workflow, a Raspberry Pi system, and robotics hardware are all relevant deployment contexts. A device category is not a guarantee that every model, board configuration, operating system, or workload supports identical performance.

Access is provided through discussions with the Vivu team and pilots. Useful initial details include the device or board, video input, intended workload, and the behavior the application needs. Contact information is available on the Contact page.

Specific supported configurations, performance measurements, integration interfaces, deployment packaging, and commercial terms are defined for the engagement. The public capability description does not establish a downloadable SDK, a universal FPS figure, an end-to-end offline guarantee, automatic cloud-index synchronization, or a published edge-model tool on the remote MCP server.

See On-device video intelligence.

Robotics VLM optimization

A vision-language model, or VLM, connects visual inputs with language. Vivu helps robotics teams improve the quality of the model's understanding of physical scenes, with a focus on two related tasks.

Object recognition

The first task is recognizing the objects that matter to the application in the scenes and operating environments the team works with. An engagement starts with the customer's model and visual requirements.

Relationships between objects

The second task is understanding how objects relate within a scene. Illustrative relationships include a cup on a table, a part inside a tray, and one object next to another.

Recognizing both a cup and a table is different from understanding that the cup is on the table. This distinction describes the focus of the work. The examples illustrate the task; they are not published accuracy benchmarks.

Engagement scope

The current emphasis is recognition and relationship-understanding quality. A specific training method, model architecture, accuracy improvement, inference acceleration, memory reduction, or cost reduction is not a general published promise.

A useful conversation with Vivu describes the VLM, object categories, relationships of interest, representative scenes, and the operating environment. Contact the team through Contact.

Recorded robot runs

Robotics teams can also use hosted video search on recorded runs. Natural-language queries retrieve candidate moments with source recordings and timecodes for incident review and human-reviewed training, evaluation, or QA datasets.

This is a separate workflow from a VLM optimization engagement. Search helps locate footage; final selection and interpretation of candidate training or evaluation examples remains a review step.

See Robotics.

Solutions and use cases

Solutions maps Vivu capabilities to team workflows. The following descriptions summarize the current scope.

Marketing

The marketing workflow starts with footage a team already owns. Uploading creates a searchable index of speech and visuals, so a query can describe the desired shot rather than guess a filename or depend on manual tagging. Teams can search before a shoot, index new footage after a shoot, pull selects during editing, and reuse older material for new work.

References to storage services in the marketing material describe where customer footage may originate. They do not by themselves establish automatic Drive or Dropbox synchronization. The documented hosted ingestion path is browser upload.

See Marketing.

Customer support

Customer clips can be indexed and queried for what is visible as well as what is said. Silent clips can still contain useful visual signals such as actions, objects, or error text. Results return candidate moments tied to their source clip, so an agent or product team can inspect the actual evidence.

Diagnosis, case outcomes, refunds, and escalation decisions stay with the support organization. The support page's workflow describes uploading clips into Vivu; it does not establish a native integration with every help-desk platform.

See Customer support.

MCP connector

Connection details

The remote server allows a compatible assistant to use the hosted product. No anonymous project access is provided. The user signs in and authorizes the connector outside the conversation.

The public site provides entry points for Claude and ChatGPT. The published custom-connector instructions use Claude as the main example; other compatible clients need support for the documented remote transport and OAuth flow. The endpoint alone does not prove that every client or account configuration supports the connector.

Sources: MCP server, Connector reference, and Application connector guide.

Setup

In Claude, the documented directory flow is Settings → Connectors → Discover, search for Vivu, and connect. The public listing is Vivu in Claude. A custom connector can instead be added using the published endpoint, followed by sign-in and enabling the tools for the conversation.

Client menus may change. The current setup pages linked above are the detailed reference for supported clients and account setup.

Documented operations

The connector's user-facing workflow includes:

• Listing and creating projects.
• Opening a project's secure upload page.
• Listing project videos and checking indexing status.
• Searching a project with a natural-language description.
• Checking an asynchronous Precise search until completion.
• Returning candidate time ranges and result-page links.
• Deleting a video or a project.

This is an operation summary, not a tool schema. The connected server's current schema determines the exact available tool names, required fields, types, and authentication requirements. File uploads occur through the browser page in the documented workflow.

Deletion is permanent. The exact target and intended deletion should be clear before the operation is executed. Deleting a project affects its videos, and there is no documented undo or restore.

Example workflow

1. Connect a compatible assistant to Vivu and complete sign-in.
2. Choose or create a project, such as Launch Interviews.
3. Open that project's upload page and upload authorized footage in the browser.
4. Check indexing status until the required videos are searchable.
5. Ask for relevant moments, specifying the project and optionally Fast or Precise mode.
6. When needed, check the existing Precise job until it completes.
7. Open the result page and verify candidate moments against the footage.
8. Prepare the selected original segments for download when required.

Example search request:

In Launch Interviews, find moments where a customer explains why they switched products. Use precise search.

Common operational problems

Detailed behavior is documented in Limits and errors.

Skills

Vivu publishes downloadable skill packages that organize complete assistant workflows around the connector. A skill supplies instructions for a task; MCP supplies access to the hosted video operations. Skills can also depend on other tools, services, user files, or review steps identified by the individual skill.

The catalog supports Claude and Claude Code workflows. Packages are free to download. Their Vivu indexing and search calls use the connected account's existing allowance; other required tools have their own access conditions.

The installation guide describes adding the Vivu connector, then uploading a skill ZIP in the Claude application or installing its folder for Claude Code. See the Skills catalog and Installation guide for current instructions and prerequisites.

Catalog categories

Representative packages include Competitor webinar battlecard, Loom recordings to runbooks, Sidewalk situation shortlists, and Robot episode label spot-check.

A package that produces a document or interacts with another application is a workflow combining tools. Its output does not imply that Vivu alone has a native integration with every named application. Individual package pages describe the current inputs, dependencies, and review steps.

Hosted pricing and allowances

The following describes the published hosted web app and connector plans as of the update date. The Pricing page is the source for current commercial details.

The Free plan has a monthly allowance rather than a one-time trial allowance. Indexing a video consumes its duration in indexing minutes once. Subsequent searches consume search credits instead of indexing that recording again. Fast search uses one credit; Precise uses five in total.

The web app and connector share the same balance. A video longer than the remaining indexing allowance may be offered partial indexing by the web app. When allowance is exhausted, the available next steps include waiting for the next billing cycle or changing the plan. Plan management is available from Profile in the application.

Teams whose needs exceed Premium, or who want to work in an organization workspace, can contact Vivu. The site lists Free and Premium as its named plans. Edge-device deployments and VLM optimization are scoped separately with the team; the table above does not price those services.

Data handling and access

This section describes the hosted video search and remote MCP service. On-device deployments are discussed separately under On-device video intelligence.

Hosted videos are uploaded into a project in the customer's Vivu account and indexed in Vivu's cloud. Access depends on the signed-in account, project or workspace permissions, and authorization of the connector. Signing in to an assistant alone does not grant access to Vivu projects.

Uploaded content remains associated with the video or project until deletion. Deleting a video or a project removes its media from Vivu's cloud and is permanent. The application is not a substitute for keeping the customer's own source copies.

The Security page identifies Google Cloud's us-central1 region in the United States for the product backend and directs readers to the Privacy Policy for retention, processing location, and providers. The current Security page, Privacy Policy, and Terms are the authoritative public references for those subjects.

Only footage the customer is authorized to share and process should be uploaded. This reference does not add a security certification, training-data policy, retention SLA, or regional deployment option beyond the published policies.

Questions that distinguish the offerings

What should an assistant use to search a customer's existing video library?

The hosted web app or authorized remote MCP connector. Videos must be in a project and have completed indexing. Search returns candidate moments associated with source recordings.

Is the optimized device model only for phones?

No. The deployment contexts include phones, Raspberry Pi systems, and robots. Hardware and workload fit are scoped with the Vivu team.

Can an assistant assume the remote MCP endpoint runs the device model?

The published remote connector reference covers hosted projects, upload links, indexing status, and search. Device-model integration details belong to the scoped deployment; they are not established by the hosted MCP workflow.

Does robotics optimization mean a faster VLM?

The current offering described here emphasizes object recognition and relationship-understanding quality. It does not establish a general speed, memory, or cost improvement.

Does a search result constitute a verified event or a finished training label?

It is a candidate moment. The source footage needs to be checked for the intended use, including selection for evaluation or training datasets.

Can a single query search all projects?

The documented hosted connector searches one project at a time. Cross-project search is not part of the published workflow.

Does downloading a skill include unlimited indexing and search?

No. Downloads are free, while connector calls use the account's existing Vivu allowance. Other workflow dependencies are described by each package.

Where should a prospective device or robotics customer start?

The Contact page, with a description of the device or model, relevant video tasks, operating environment, and desired outcome.

Website resource map

Product and company pages

Guide topics

The Guide contains educational articles and comparisons about working with existing video. An article about an external tool or an editing technique is not a declaration that Vivu provides that tool's capabilities.

The topic and skill catalog pages link to the individual articles and packages. They are the current inventories as those collections grow.

Agent-readable versions and contact

• Product reference: the HTML rendering of this reference.
• Complete Markdown reference: the same product content in plain text Markdown.
• Short discovery index: a compact overview with links into the reference and site.
• Contact the Vivu team: product evaluation, larger hosted needs, on-device pilots, and VLM optimization.
• Support email: product support.

Public product pages describe the offering; the connected server's current schemas describe executable MCP operations; published policies govern their respective subjects. Pricing and operational details should be checked against the current linked source when a decision depends on them.