Workflows

From capture to delivery,
without the waiting.

Every discipline below has the same enemy: copying. Uploads, downloads, hand-offs, "which version is this" — all of it is copying wearing different hats. Stone mounts as a drive and makes the copy unnecessary. Here's what that does to a working day.

Video

Shot at 10:14. In the timeline by 10:15.

Ingest on set and the folder in town fills as the cards drain — not a copy of the footage, the footage. Camera originals play the moment they land: scrub an 80 GB original like a proxy, because Stone streams the frames you touch and skips the ones you don't.

Proxies appear in place

Spin up a cloud container that mounts the filespace right next to the footage. It watches the ingest folder and encodes as cards arrive. Your NLE doesn't run a pipeline — it just sees proxies show up where proxies belong.

Two cities, one folder

The editor in Oslo and the colorist in LA aren't syncing two folders. They're in the same one. A new grade, a new cut, a relinked reel — visible to the other the instant it exists. Live is not a feature here; it's the default.

on-set — ingest station
$ stone run encode-proxies --watch ~/Stone/Ingest/CardA --profile prores-proxy
 container up — mounted beside the footage
 watch: Ingest/CardA/A001C007.mxf  (new, 41.8 GB)
  encoding… done in 3m 12s → Proxies/A001C007.mov
# the edit suite never noticed a thing. it just got proxies.
Audio

The session never leaves. People just join it.

Record multitrack straight into the filespace from the live room — it's fast enough to be the record drive. From that moment the session is already everywhere it needs to be, and nobody ever builds a zip again.

Track

Arm 32 inputs and print them direct to the cloud. No scratch disk, no 2 a.m. offload, no "did we back up the drums."

Bounce

Bounce the session — the producer gets the notification before you've left the live room. No upload, no link, no waiting.

Hand off

Share the session folder — the whole tree, as it sits. No zips, no missing audio files, no "reply-all with the stems."

Recall

Every take and every bounce is kept, to the nanosecond. When the client suddenly loves the rough mix again, rewind to it.

Photography

A million RAWs, one shelf.

A wedding season is 400,000 frames, three card backups each, and a folder named final_final_v2. Stone stores each chunk of it once — the bedrock doesn't care how many times you dragged it in.

Ingest that dedups

Re-imports, card backups, and backup-of-the-backup folders collapse into one stored copy. You pay for the light you caught, not your filing habits.

Versions, not copies

Selects and retouches ride on the file as history, not as TIFF number nine. One filename, every state it's ever been in, rewindable.

Shoots that cool off

Old shoots slide to archive and long-term tiers on their own. Hot at delivery, cold by winter — same folder, still instant to browse.

Galleries in a link

Share a folder as a read-only client gallery the second the edit is done. Permission-controlled, live immediately, revocable just as fast.

VFX & animation

Render farms that mount, not copy.

Staging assets onto a farm and staging frames back off is a job nobody billed for. Cloud render nodes mount the filespace right where the data lives — nothing shipped in, nothing shipped out.

farm — orchestration
$ kubectl scale deploy/render --replicas=64
deployment.apps/render scaled
$ stone status ~/Stone/Shots/ep104
 64 render nodes mounted · zero bytes staged
 renders/sc12_v031/  frame 0481 … 0482 … 0483
# the review room sees each frame the second it lands.

Frames land in place

Every node writes into the shot folder itself, and every workstation on the show sees each frame the instant it exists. The supervisor is checking frame 0481 while frame 0482 renders. Dailies stop being daily.

Scale is a number, not a project

Sixty-four nodes tonight, four tomorrow. Containers mount the filespace like any other client, so the farm is just compute — the data never moves.

Kubernetes Container images Bare metal Any renderer that writes files
AI pipelines

Point your models at the bedrock.

Transcription, tagging, scene detection, upscaling — AI workloads read and write the filespace directly, under the same permission control as everyone else. Connect your AI, keep your permissions.

Reads what it may

A model sees exactly what its credentials allow — down to the folder. Nothing gets exported to a side bucket and quietly forgotten there.

Writes beside the source

Transcripts next to the interviews. Tags next to the footage. Upscales next to the originals. Results are ordinary files, in the folder where the team already works, visible the moment they land.

Runs where the data is

Spin the workload up in a cloud container that mounts the filespace, or drive it over the REST API. Either way, nobody shuttles terabytes to a GPU.

Architecture & engineering

One model. Every consultant. No "which version?"

Big BIM and CAD models are gigabytes of geometry leaning on shared reference libraries — exactly the shape deduplication loves. The libraries are stored once, however many projects and firms lean on them, and everyone opens the current model. Not Tuesday's.

Cross-firm, without the portal

Share the project folder with the structural engineer's company and it's in their filespace instantly — permission-controlled, revocable, live. When they save, you see it. No upload portal, no Friday sync, no courier drive.

The audit trail is the filesystem

Every change is kept with nanosecond granularity. "Who moved the west stair core, and when" is a question with an exact answer — and a rewind button.

project — Fjordgate
$ stone history ~/Stone/Fjordgate/Fjordgate-C3.ifc --today
14:02:11.482031554  maja@span.no      wrote 38 MB   structural, level 3
13:47:03.114820019  omar@atelier.no   wrote 2 MB    façade panels, west
09:15:44.902113407  you               opened        read only
# who changed what, when — to the nanosecond. arguments end early here.
Surveillance & continuous capture

Footage that files itself.

Cameras write straight into the filespace, and the footage manages its own retirement: hot, standard, archive, long-term — one filespace, four temperatures, zero migration scripts.

Capture

Cameras and encoders write directly in, around the clock. No middlebox, no export ritual, no drive that fills on a holiday weekend.

Serve

Recent footage stays hot. Incident review scrubs like local disk, and watch notifications feed your alerting the moment new footage lands.

Cool

Per your retention policy, footage steps down to archive, then long-term. The cost falls; the folder doesn't change.

Prove

History is immutable, timestamped to the nanosecond. When footage becomes evidence, its integrity is the system's word — not a chain of exported copies.

The open API

REST over everything.

Everything the client does, the API does. Automate ingest, build a review portal, wire delivery into whatever you ship with — it's all files, and it's all one call away.

Watch a folder over the API and your systems learn about new files at the same instant your people do. Delivery portals that update themselves, ingest that triggers reviews, archives that confirm receipt — small scripts, big manners.

Read the API documentation →

anywhere — curl
$ curl -N -H "Authorization: Bearer sk_live_…" \
    "https://api.stone.no/v1/spaces/northlight/files?watch=Delivery"
event: file.created
data:  { "path": "Delivery/EP104_v12.mov", "size": 84523118592 }
# your delivery portal knew before your producer did.

Your workflow isn't on the list?
It will be.

Stone is in invite-only preview. Tell us what you make, and we'll make room.