Jumper
Jumper is a machine-learning search engine for video footage that runs entirely on your own machine. It finds shots, spoken words and faces across hours of rushes, inside Premiere Pro, Final Cut Pro, DaVinci Resolve and Avid.
What is Jumper?
Jumper is a machine-learning search engine for video footage, built by the Swedish company Witchcraft Software AB. It runs as a standalone desktop application on macOS and Windows and as a plugin inside Adobe Premiere Pro, Final Cut Pro, DaVinci Resolve and Avid Media Composer, so that editors can search without leaving the timeline.
The tool indexes footage locally, then answers three kinds of query. Visual search retrieves clips from plain keywords with no manual tagging, the index being built by the analysis pass itself. Speech search finds a spoken word or phrase across hours of interviews in 111 transcription languages, with dedicated models for Japanese, Swedish and Moroccan Arabic. Face recognition and People search, reserved for the Pro tier, detect and index every person on screen, so that typing a name or clicking a face returns every appearance. Companion tools refine the results: Match Similar surfaces visually related shots, while Match Source Monitor, Match Timeline and Match Pasteboard frame search from a reference image to within roughly one second, since analysis samples one frame per second.
Two newer modules extend the workflow. The Script Editor opens a full transcript in a single workspace, allows non-destructive editing, silence removal and line trimming, then sends selected ranges to the sequence or exports them for review or captions. Agentic Editing exposes a local MCP server, again on the Pro tier, with one-click connection for Claude Desktop, Claude Code and OpenAI Codex Desktop; the agent only ever receives metadata such as file paths, timecodes, transcript excerpts and result lists.
Everything runs on the machine. Footage is never uploaded, and once the licence is activated Jumper works entirely offline, paid licences revalidating weekly while trial licences need a permanent connection. The publisher states that it does not train AI on customer media and keeps no data from it. Claimed performance is roughly 100 hours of footage searched in about 0.2 seconds, with analysis files weighing around 0.1% of the source media. The company was founded in Sweden in 2024 by a professional editor working with two engineers, and shipped its first version that November.
What it does
- Search any shot by keyword, with no prior tagging
- Find a word or a phrase spoken anywhere in hours of interviews
- Locate every appearance of a given person across an entire library (Pro tier)
- Surface visually similar shots from a clip you have already picked
- Search for footage matching the frame held in your source monitor
- Work a full transcript and send the selected ranges straight to the timeline
- Let an AI agent run searches and prepare a sequence through MCP (Pro tier)
When to use Jumper / When not to
A quick filter to help you decide if Jumper is the right fit.
When to use Jumper
- Documentary editors who need to locate every appearance of a contributor across hours of rushes
- Reality TV and unscripted teams assembling confessionals, reactions and confrontation scenes by cast member
- Corporate and commercial editors pulling testimonials, executive soundbites and campaign B-roll
- Post-production teams working under NDA, since the footage never leaves the machine
- Editors working offline or on unreliable connections, and freelancers who prefer a perpetual licence to a subscription
When not to use Jumper
- Anyone expecting a mobile or web version: Jumper is a desktop application only
- Editors who need to search technical metadata such as frame rate or camera model, or folder names, both still on the roadmap
- Teams that want their folder hierarchy mirrored inside the tool, which it does not reproduce
- Users looking for an editing or generation tool: Jumper finds and prepares selects, it does not replace the NLE
- Buyers who need a permanently free tier, or several active machines on a single licence key
How to use Jumper
A typical end-to-end flow, from setup to results.
- Download the desktop application for macOS or Windows from the website
- Activate a free trial key or a paid licence key; trial keys require a permanent internet connection
- Install the plugin for each editing application you work in
- On macOS with Final Cut Pro only, grant Accessibility, Screen Recording and Full Disk Access
- Run the analysis on your media, or set a Watch Folder so new files are analysed automatically
- Search by visual keyword, by spoken word, or by face on the Pro tier
- Refine the results with Match Similar, Match Source Monitor, Match Timeline or Match Pasteboard frame
- Save what you keep into personal or shared tag collections
- Open the Script Editor to work the transcript and push selected ranges to the timeline
- Install the MCP connection from the Settings tab, AI Agents section, to drive Jumper from an agent (Pro tier)
Pros & Cons
Pros
- Fully local processing: the footage never leaves the machine, which suits NDA work
- Works offline after activation, paid licences needing only a weekly revalidation
- The publisher states explicitly that it does not train AI on customer media and keeps nothing from it
- Integrated inside the four major editing suites rather than sitting beside them
- Search speed given as roughly 100 hours of footage in about 0.2 seconds, with no manual tagging
- Perpetual licences are available alongside subscriptions, each including a year of updates
- Public API and local MCP server allow automation without leaving the workstation
Cons
- No permanent free plan, and the trial is capped at 40 searches rather than a duration
- One active machine per licence key, and any workstation reading shared analysis data needs a Pro licence
- The most distinctive features, faces, agent, API and sharing, are all confined to the Pro tier
- No mention of the GDPR anywhere, no DPA, no Article 27 representative and no dedicated privacy contact
- Pricing figures are rendered in JavaScript and never appear in the source of the pricing page
- Final Cut Pro users must grant three broad macOS permissions, including Full Disk Access
- No mobile application and no web version; metadata and folder-name search are still on the roadmap
Pricing & Plans
There is no permanent free plan. A free trial is available without a credit card, allowing unlimited media analysis but capped at 40 searches. The lowest paid entry point is Jumper Pro at USD 49.00 per seat and per month; the standard Jumper tier is sold only as a perpetual licence at USD 169.00 per seat, while Jumper Pro is also offered at USD 399.00 per year or USD 599.00 perpetual. Perpetual licences include one year of updates, after which a further year may be purchased at 50% of the prevailing perpetual price. Payments are handled by Paddle, and a 14-day refund window applies to every plan. These amounts come from the publisher's official announcement of 3 October 2025, the pricing page rendering its figures in JavaScript.
- USD 0 - unlimited media analysis but capped at 40 searches
- requires a permanent internet connection
- Jumper Standalone
- plugins for every supported NLE
- unlimited analysis
- unlimited visual and speech search
- Script Editor
- search in any media outside the project
- one device per licence
- everything in Jumper plus shared analysis files between devices
- Swarm analysis
- access to all AI models
- facial recognition
- People search
- Agentic Editing and the Jumper API
Data, GDPR & hosting
A consolidated view of how Jumper handles your data.
GDPR overview
There is no GDPR implementation to report: the acronym appears nowhere on the website or in the documentation. No Article 27 representative is designated, no Data Processing Agreement is published or offered on request, no data protection officer is named, and no dedicated privacy or legal address exists, hello@getjumper.io being the only one published. The privacy policy, last modified on 28 January 2026, describes what the software does and names seven third-party providers, but sets out neither data subject rights, nor retention periods, nor a legal basis for processing. The EULA is governed by Swedish law, with exclusive jurisdiction in the Swedish courts. Witchcraft Software AB is established in Sweden and the regulation therefore applies to it by law; what is recorded here is the silence of the documentation, not a judgement on the vendor's practice.
Who owns the data?
Under the EULA the software is licensed, not sold, and Witchcraft Software AB retains every right in it. Everything else belongs to the user: the media files, the analysis files Jumper generates such as indexes, thumbnails and transcriptions, and all cache data. The publisher states that it never takes custody of footage, and access to files is governed by the operating system's own permissions, exactly as for any other application. In shared setups, indexes and cache stay on the customer's own servers and network. Nothing related to the user or the footage leaves the computer apart from licence verification and crash reports.
Reuse rights
Users keep full freedom over what Jumper produces. Analysis files, indexes, thumbnails and transcriptions are written to their own disk and can be copied, moved or transferred to another editor's machine without asking permission; the documentation explains how to relocate the cache folder. Selected ranges can be pushed to the timeline or exported in several formats for review or captions. The EULA restricts the software itself, not the results: reselling, renting, redistributing or reverse engineering Jumper is forbidden, as is using technical analysis of it to build a competing product. On the vendor side, an internet connection is needed only for licence verification. The email address serves for authentication and licence-key access, with an optional mailing list that can be left at any time, and crash reports go to Sentry and can be disabled. Named third-party providers are Keygen.sh and Google Cloud for licensing, Paddle for payments, Sentry for crash reporting, Tolt for affiliate links, GitHub for documentation hosting and Cloudflare for the website.
Data retention & training
Hosting summary
There is no hosting of user content to speak of, and that is the product's central claim. Media files stay wherever the user put them, on local drives, a NAS or the customer's own servers, and every analysis pass, speech transcription, face detection and visual indexing runs on the user's machine or local network. Jumper only writes metadata files, indexes, thumbnails and transcriptions, stored locally beside the media. In team environments, indexes and cache remain on the customer's own infrastructure. The vendor declares no hosting country and no hosting region, which is consistent with this design. Remote services are limited to licence verification through Keygen.sh and Google Cloud, payments through Paddle, crash reporting through Sentry and affiliate links through Tolt. The marketing site is hosted by Cloudflare and the documentation on GitHub through Mintlify. The domain resolves to a United States address, 104.18.16.136, but this is a Cloudflare anycast node serving the website and says nothing about the location of customer data.
Things to keep in mind
Risks and trade-offs to weigh before adopting Jumper.
- Check the prices before committing: the pricing page renders its figures in JavaScript, and the amounts recorded here come from an official announcement dated 3 October 2025
- The publisher is established in Sweden and therefore bound by the GDPR, yet the site never mentions it, publishes no DPA and names no Article 27 representative or data protection officer
- Every legal, privacy and support matter goes through the single address hello@getjumper.io, with no dedicated legal or privacy contact
- Full Disk Access, Screen Recording and Accessibility are broad macOS permissions; granting them for Final Cut Pro widens what a compromised application could reach
- Crash reports are sent to Sentry by default, and disabling them is documented but is not the initial setting
- Team costs escalate quickly, since every machine that merely reads shared analysis data requires its own Pro licence
- Leaning on automated search can quietly erode an editor's own familiarity with the rushes: the tool returns pointers, and the judgement about what belongs in the cut stays human
Setup & Integrations
Technical difficulty
Installation is straightforward: download the desktop application for macOS or Windows, activate the licence key received by email, then install the plugin for each editing application used. Friction is concentrated on macOS with Final Cut Pro, which requires Accessibility, Screen Recording and Full Disk Access; none of these are needed with Premiere, Resolve or Avid. DaVinci Resolve must stay in English with its default shortcuts, and Avid must keep its default I, O and Q keys. The firewall must allow api.keygen.sh. Expect an initial analysis pass, where an Nvidia GPU or Apple Silicon is far preferable to CPU alone.
Deployment
Integrations
Supported languages
Behind Jumper
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
Which editing applications does Jumper work with?
Does my footage go to the cloud?
Does Jumper train AI on my media?
Can I work offline?
How many machines does one licence cover?
What does the free trial include?
How many languages does speech search cover?
Is there an API?
What is the refund policy?
How much disk space do analysis files take?
Should you pick Jumper?
Jumper occupies a deliberately narrow position: it searches footage and prepares selects, and leaves the cutting to the editor. Within that scope it is unusually coherent. The local-processing claim is not a line bolted onto a cloud product, it runs through the documentation, the licensing model and the agent integration alike, and it is what makes the tool usable on NDA material or on a workstation with no reliable connection.
The commercial model is equally unusual for an AI product. Perpetual licences sit alongside subscriptions, at USD 169.00 for the standard tier and USD 599.00 for Pro, each including a year of updates. The entry cost is real for an independent editor, but nothing forces a recurring commitment and the 14-day refund window covers every plan.
Two reservations deserve weight. The first is documentary rather than technical: for a company established in Sweden, and therefore squarely within the scope of the GDPR, the complete absence of the acronym from the site, with no DPA, no Article 27 representative and no privacy contact beyond a single generic address, is a gap a corporate buyer will notice whatever the underlying practice. The second is commercial: face recognition, agentic editing, the public API and shared analysis all sit behind the Pro tier, and every machine that merely reads shared analysis data needs its own Pro licence, so team budgets rise faster than the headline price suggests.
The company is young, founded in 2024, but the product is already integrated into the four major editing suites. For editors who spend their days hunting through rushes and would rather that footage stayed on their own disk, Jumper answers a specific and expensive problem well. Anyone needing mobile access, metadata search or a free tier should look elsewhere.
- Choosing a selection results in a full page refresh.
- Opens in a new window.