Products
What we've shipped, and what's next.
Oragonlabs is a research lab, and this is our agenda in the open. Voiceit is the first product built on our engine, and the Oragon SDK puts that engine in developers' hands. Each programme after it is an area where we believe sovereign, efficient AI changes what an institution can do, and where we are actively doing the research.
Voiceit. Just say it. We'll write it.
Our first product: a dictation and voice-notes app. Hold a shortcut in any app, speak naturally, and clean written text lands at your cursor.
Keep what matters as notes with automatic titles and full search. Voiceit runs on the engine we build at Oragonlabs, and it is where that research meets everyday use.
- Dictation:
- Any app with a text field, no plugins
- Notes:
- Automatic titles, full search
- Languages:
- Choose yours; teach it your names and terms
- Pricing:
- Start for free
- Status:
- Live
Oragon SDK. Run AI models inside your own software.
The engine behind Voiceit, as a Python package. On a laptop, a server or a datacenter, CPU or GPU, with your data never leaving the machine.
Validated structured output, honest decisions, tool use, cited answers over your documents and an OpenAI-compatible API, with typed errors that say whether to retry.
- Install:
- pip install oragon. No API key
- Runs on:
- macOS (Apple silicon), Linux x86-64 and arm64, CPU or GPU
- Language:
- Python 3.9 to 3.13; any OpenAI client over HTTP
- Pricing:
- Free up to 4 concurrent requests, commercial use included
- Status:
- Beta
Apart from Voiceit and the Oragon SDK, none of the programmes below are commercially available yet. They are active research directions. We publish them so partners can shape them early.
Research programmes
Six directions, one thesis.
In every case the question is the same: can this run close to the people it serves, on hardware they control, at a price their economy sustains?
Intelligent video systems
Video understanding that runs on the cameras and servers an organisation already owns, detecting events rather than streaming footage to somebody else's cloud.
Most video analytics today require sending continuous footage off-site. We are researching on-device inference that keeps raw video inside the perimeter and emits only the events an operator asked for, so a site gains awareness without creating a new surveillance liability.
Shape this programme- Where it runs:
- On-premise, on existing camera infrastructure
- Data leaving site:
- Events only, not raw footage
- Connectivity:
- Designed to operate offline
- Status:
- [research direction]
Assistive technology
AI that widens access for people with disabilities: speech, vision and language assistance that works on affordable devices, offline, in local languages.
Assistive AI is overwhelmingly built for high-resource languages and expensive hardware. We are researching small models for speech-to-text, text-to-speech, scene description and communication assistance that run on the phones people in our markets actually carry.
Shape this programme- Where it runs:
- On-device, mid-range and older phones
- Connectivity:
- Full offline operation
- Languages:
- [local languages: confirm]
- Status:
- [research direction]
Identity & civil registration
Identity verification and civil registration infrastructure that runs on national systems, under national law, owned by the institution accountable for it.
Identity systems are among the most sensitive a state operates. Our research position is that they should run on sovereign infrastructure with auditable data handling, never as a rented service from a foreign provider whose terms can change.
Shape this programme- Where it runs:
- Sovereign / national infrastructure
- Ownership:
- Operated by the accountable institution
- Auditability:
- Logged, inspectable, exportable
- Status:
- [research direction]
Defence & public safety
Decision-support and situational-awareness systems for national security institutions, built to work in denied, degraded and disconnected conditions.
Defence workloads are where the sovereignty argument is least negotiable: no capability that a foreign provider can switch off is a capability at all. Our research focus is resilient, self-hosted systems that keep functioning without external connectivity.
Shape this programme- Where it runs:
- Self-hosted, air-gap capable
- Connectivity:
- Denied / degraded environments
- Dependencies:
- No foreign provider in the loop
- Status:
- [research direction]
Clinical decision support
Diagnostic and triage support that runs inside the hospital, so no patient record leaves the building and the system survives a network outage.
Health data is the hardest case for cross-border AI and the highest-value one for patients. We are researching compact clinical models that a hospital can host on its own hardware and keep running when connectivity fails.
Shape this programme- Where it runs:
- On-premise, hospital-owned hardware
- Data residency:
- Never leaves the building
- Connectivity:
- Full offline operation
- Status:
- [research direction]
Financial infrastructure
AI infrastructure that satisfies local data-residency law by construction and is priced against local transaction economics.
Financial institutions here are asked to choose between compliance and capability. We are researching in-country deployable infrastructure that removes that trade-off, running on hardware institutions already own rather than in a distant cloud.
Shape this programme- Where it runs:
- In-country / private cloud
- Compliance:
- Local data-protection law by design
- Tuning:
- Local hardware and operating conditions
- Status:
- [research direction]
How we decide what to build
Sovereignty cuts both ways.
Some of the systems above (video understanding, identity, defence) are powerful enough to harm the people they are meant to protect if deployed carelessly. The same principles that make them sovereign make them accountable: data that stays in jurisdiction stays subject to that jurisdiction's law, and systems that are auditable can be audited. We publish our data-handling terms, we design for logging and inspection, and we take a position on deployments we will not support.
[Our data-handling and deployment policy is being finalised and will be published here.]
Shape a programme before it ships.
We build these with institutions, not for them. If one of these directions is your problem, we want to hear how you'd use it.
Get in touch: hello@oragonlabs.com