Embed transcription, semantic search, clip production and NLE interchange as a signed binary — scoped by a Grant, licensed on flexible terms. Your users' media never leaves their machine.
Signed binary · localhost /api/v1 · no cloud dependency
The pipeline
The engine ingests media, understands it, and produces from it, without a single byte leaving the machine. Embed the whole path, or just the stage you're missing.
Video, audio, images — the files your users already have.
Transcribe, faces, OCR, logos, actions, scenes, music.
Find the exact moment by meaning, with an explainable trace.
Render a clip, or emit FCPXML / EDL / OTIO.
Results key back by Moment ID with file handoff.
Every stage runs behind localhost /api/v1 — nothing uploads, and the network can be off.
Components
The engine is modular. License the capability that fills your gap — enabled from the same binary by your Grant — and add more as you grow.
● Available
Turn any moment into a finished vertical clip inside your app — auto-reframe to 9:16, burned-in captions, highlight selection and render. No cloud round-trip.
For creator and clip tools, social schedulers, CapCut-style editors.
Get started● Select partners
Transcription plus semantic search across everything in a file — words, faces, OCR, logos, actions, scenes and music — indexed on-device with an explainable retrieval trace.
For DAM/MAM, podcast and meeting tools, archives.
Contact us● On request
Clean FCPXML, EDL and OTIO with drop-frame timecode for real timeline handoff — plus on-device attestation, integrity hashing and a deterministic tool ledger.
For editor vendors, post tools, legal, evidence and enterprise.
Contact salesWhy on-device
The compute is already sitting on your users' machines. Embedding the engine turns that into your feature — with properties a cloud API structurally can't offer.
Indexing, search and rendering all run locally — "we never see your media" becomes an architecture claim, not a promise.
Compute runs on the device you don't pay for. Your gross margin stops tracking users' media volume.
A signed key enables exactly the components you license and denies the rest — from one identical binary.
A flat per-integration license by default — or usage-metered, per-seat, or revenue share. Priced to your model.
Words, faces, on-screen text, logos, actions, scenes and music — one engine, indexed together.
It ships as a signed binary. No model serving, no GPU fleet, no eval pipeline to staff.
Integration
The engine runs as a local service on the platforms your product already targets. You ship the binary and your Grant; you call /api/v1 from your app.
Bundle the engine with your app. It runs as a local service exposing /api/v1 on loopback — no external host, no account for your users.
The signed Grant scopes the capability subgraph. A clip-production Grant lights up create.* and leaves search retrieval denied.
Drive it over HTTP. Results key back by Moment ID, with file handoff for rendered artifacts — so it composes with your data model.
The engine counts calls locally and reports usage — under metered terms you pay for what it does, or it's simply visibility under a flat or per-seat license.
Licensing
Because the compute runs on your users' devices, we're not passing a GPU bill through to you — so we can shape licensing around your model. Most integrations start with a flat per-integration license; usage, per-seat and revenue-share terms are on the table too.
One predictable fee for the component, regardless of volume — the simplest way to start a first integration.
Per clip rendered or file indexed. Cost tracks the value the engine delivers.
Aligns with your own model if your product already bills by seat.
Lower fixed cost, shared upside — we're invested in the deal working.
Proven in production
Faring is the engine already powering MediaFind, a released on-device media search & production app, and a family of sibling products built on the same core. We license that proven pipeline to product teams — we're onboarding the first integration partners now.
Add finished-clip production without a render backend.
Make stored media searchable by what's inside it.
Hand moments off to a real NLE timeline.
Prove processing ran on-device, with an audit trail.
Honest scope
We're deliberate about what we are — and clear about what we're not — so the fit is obvious before you invest engineering time.
FAQ
We're onboarding a small number of design partners for the first integrations. Tell us the gap you're filling and we'll scope a Grant for it.