Platform
The best AI, inside your own environment.
Any model. Your data stays under your control. It gets better the longer you use it. More than just a legal research tool.
How it works
The model comes to the data. You decide what leaves.
Irys sits between your institution and the models, inside your perimeter. It decides what each model sees, records what the work taught it, and waits for approval before anything moves.
Reads, reasons, checks, waits for approval, and remembers. Every job leaves state behind for the next one.
Four controls
Every question a risk committee asks, answered by the system.
Where it can go
With local models, processing stays inside your perimeter. If you approve an external model, Irys calls it only on your terms, and you choose which work may use it.
What it knows
Persistent state your institution owns: sources, facts and relationships, decisions and approvals, work completed, known gaps.
How it got there
Each step carries its provenance: what was read, what was decided and on which source, so a reviewer can follow the path.
What it may do
Permissions decide what each agent can see, remember and do. Actions wait at the approval gate.
The approval gate
Nothing executes until it is approved.
Every job moves through the same five steps, in the same order, inside your environment.
- 01ReadReads authoritative sources and your records, within its permissions.
- 02Update stateRecords what it learned: entities, relationships, decisions, open questions.
- 03ProposeDrafts the change or the answer, with its sources attached.
- 04VerifyChecks the proposal against its sources and your rules.
- 05ExecuteActs only once the step is approved.
Why it compounds
Rent the model. Own the memory.
Models change every few months. What your work teaches the system stays in your perimeter, in a persistent state you own, so each job starts further ahead than the last.
- Reads authoritative sources
- Resolves entities and relationships
- Records decisions and open questions
- Sources
- Facts and relationships
- Decisions and approvals
- Work completed
- Known gaps
- Starts with accumulated context
- Reviews what changed
- Reuses reviewed decisions and controls
Deployment
Your environment, your choice of model.
Three ways to run Irys. The same core in each, and the model can change without the institution changing anything else.
| Deployment | Air-gapped | On-prem | Your VPC |
|---|---|---|---|
| Where Irys runs | Inside your network, disconnected | Your data centre, your hardware | Your own cloud account |
| Models | Local models | Local, or approved external | Local, or approved external |
| Calls leaving the perimeter | None | Only to models you approve | Only to models you approve |
| Training on your data | Never | Never | Never |
| Changing the model | When a better one ships | When a better one ships | When a better one ships |
Proof
Better answers from cheaper models, because of the system around them.
Irys on a low-cost model, against the legal AI leader's own agent, on the leader's own public benchmark. 2,010 tasks, $4.64 per task.
To scope and deploy inside an institution's own environment.
Every job makes the next one better. At a fintech deployment, review effort per matter is down 90%.
Strict all-pass: a task counts only when every rubric criterion passes, with no partial credit. Legal Agent Benchmark v1.0. Irys ran the full public set of 2,010 tasks across 27 practice areas on a low-cost off-the-shelf model, with no fine-tuning and every task starting from empty state. The benchmark owner reports its own agent on a private held-out split of roughly 1,200 tasks, so the comparison is across the same benchmark family rather than identical task instances. Methodology, caveats and every output are public on GitHub. See the repository.
Two ways in
One core, two doors.
The Irys app
Research, draft, review and manage legal work in one connected workspace. Start with the application and its built-in workflows.
- Ready to use
- Legal workflows
- Published pricing
Custom deployments
An enterprise AI platform that executes work using your company’s knowledge, policies and data. Connect Irys to your existing systems and configure it around your workflows.
- Your environment
- Your workflows
- Pricing agreed to scope
Same core. Start with the app, or have Irys deployed around the rest of the business.
FAQ
Frequently asked questions about running Irys in your own environment.
Which AI models can Irys run?
Any of them. Irys sits between your institution and the models rather than being tied to one, so the model can change without the institution changing anything else. The same core runs in every deployment option.
Where does Irys run?
Inside your perimeter. Three options: air-gapped, on-premise, or your own VPC. The model comes to the data, and you decide what leaves.
Is our data used to train AI models?
No. Your data is not used to train, fine-tune, or improve any AI model. Irys is covered by SOC 2 Type II and ISO/IEC 27001:2022, and data is encrypted in transit and at rest.
What does Irys add on top of a model API?
Four controls, an approval gate and persistent state. Every job moves through the same five steps in the same order, nothing executes until it is approved, and what the work teaches the system stays in your perimeter instead of in the model.
See it inside your own environment.
Bring one real workflow. We will scope it with your team and show you what a deployment looks like.
