About
An offline AIvideo player.
Built to keep media processing on your machine. Your files are not uploaded for AI processing.
Orieon is a desktop video player for Windows with transcription, translation, AI search and summarisation built in, every one of them running locally with models stored on your device and processed by your CPU or supported GPU. There is no cloud media-processing step, and no frame of your media ever leaves the device.
It is made for people who work with footage they cannot upload: researchers, legal professionals, journalists, educators, and teams sitting on material under NDA. If a recording cannot leave the building, it should not have to leave your laptop either.
Local AI processing
Transcription, translation, chat and summarisation run on your hardware. Account access, licensing, updates and model downloads may still use the network.
Private by default
Video and audio never leave the device. Nothing is uploaded to produce a transcript or a translation.
Hardware aware
A first-run benchmark measures your CPU and GPU, then picks the fastest model the machine can actually carry.
Where it came from
Two years.Three rewrites.
Apr 2024
v1.0.0
ZestSync Player
A VLC plugin that generated subtitles in the background with OpenAI Whisper. Held together with hope, never meant to be a product.
Sep 2025
v2.0.0 / v2.1.0
Standalone PyQt6 app
Cut loose from VLC and rewritten as its own PyQt6 desktop app on faster-whisper with 14-language translation. v2.1.0 brought full CUDA acceleration, 5 to 10× faster processing and proper subtitle styling.
Mar 2026
v3.0.0
Orieon
The old codebase stopped earning its keep, so the whole thing was rebuilt for offline AI from the ground up: a native Tauri and Rust shell with a C++ inference engine underneath. New stack, new engine, new backend, new licensing. Nothing carried over.
Who's behind this
One person.No team. No funding.
Orieon is built by me, Anurag, a solo developer. The desktop app, the AI engine, the backend and this website are all built and maintained by me. I got tired of "AI video tools" that quietly mean uploading your footage to somebody else's server.
I would rather ship something slower and genuinely private than something fast that depends on the cloud in the background. A bug report, a feature request or a complaint reaches me directly, not a ticket queue.
What I bring
Full-stack product engineering
React, TypeScript, C++ and Python. The interface, the systems layer and everything wiring them together, written in one place, with no handoffs.
On-device AI & inference
Local inference across C++ and supported CUDA, Vulkan and CPU backends, built to keep media processing on the device.
LLM & retrieval
Retrieval pipelines, local language models and semantic search grounded in local content: the layer behind Orin, Orieon’s AI assistant.
Backend & infrastructure
APIs, relational data, authentication, licensing, payments and administration tooling built as one connected product.
Desktop app development
Tauri for the shell, C++ for the engine and React for the interface, packaged as a Windows desktop application.
Broader technical experience
Languages
Machine learning
Vision
On-device inference
Generative AI
Backend & data
Platform
Shell & tooling
Available for hire
I built this whole product alone. Picture it with a team.
What I'm looking for
Why it's built this way
Shipping AI that calls an API is the easy path.
Running the whole thing on a user's own hardware is a harder problem, and one most products never bother solving. This one does.
01
Offline inference is the hard part
Bolt a cloud API onto a player and you can ship within a week. Making transcription, translation and a local language model run quickly on an ordinary laptop is a different problem entirely, and it is the one this project chose.
02
Three independent layers
Licensing is checked in the React front end, the Rust shell and the C++ engine, each verifying on its own. Getting past one does not open the other two.
03
Nothing billed per request
Inference happens on your machine, so there is no compute bill that grows with usage. Those economics are what allow the Premium tier to remain affordable.
What's inside
Five tools.One player.
Transcription
Whisper models turn speech into timed subtitles on your CPU or GPU, across 99 languages.
Translation
Render any transcript into 21 target languages with processing performed on the device.
Orin, Orieon’s AI assistant
Ask questions about the video in front of you and get answers pointed at the exact moment.
Summary
Structured highlights with navigable timestamps pulled out of long recordings.
Playback
Subtitle styling, speed and resume behaviour tuned to how you actually watch.
Who it's for
Footage thatcannot leave.
The people who need subtitles most are usually the ones least permitted to upload the file. That constraint comes first here, which is why there is no upload path to begin with.
Research interviews
Hours of recorded conversation turned into searchable text for coding and quotation, with consent never resting on a third-party server.
Legal review
Depositions, hearings and client calls searched by phrase and timestamp instead of scrubbed by hand.
Newsroom material
Raw footage and rough cuts read as text before the story is written, on a machine that stays in the building.
Lecture capture
Recorded classes captioned, translated and searchable for the students who need them, without moving the file.
NDA teams
Footage under agreement never leaves the drive it arrived on. Nothing to disclose to a vendor, nothing to log.
Archive footage
Old tape and unlabelled reels catalogued by what is actually said in them.