For investors
The system of intelligence.
Starting with legal. Irys turns an institution's context into finished, verifiable work, using the best model for each job, inside the institution's own environment.
Why now
Frontier models are capable. Most institutions still can't put them to work.
The hard part is everything around the model: the institution's data, its permissions, its systems, its workflows, and checking the work before it goes out. That is the layer Irys builds.
Architecture
Irys sits between institutions and their models.
When a better model ships, the institution switches to it without rebuilding anything. Irys carries the context, permissions, memory, workflow and checks from one piece of work to the next.
Thesis
Every piece of work makes the next one better.
The model is an input, not the product.
Every completed workflow leaves behind context: what happened, what changed, what was approved, what worked and what the institution knows.
Irys carries that forward. Better models make Irys better; they don't replace it.
Proof
Irys is already in production.
Performance
Irys outperformed a specialized, post-trained legal model on that model's own public benchmark, using cheaper general-purpose models.
Production
Used in live workflows across law firms, enterprises and regulated organizations, by 300+ legal teams.
Deployment
Cloud, private and customer-controlled deployment, with the models underneath interchangeable.
What customers say
“It's become our system of record.”
“The only platform I trust for real legal analysis. It actually engages in the analysis with me.”
“It understands how we work. Real, nuanced drafts, suggestions, strategy.”
Structure
The app and custom deployments, on one platform.
The Irys app
Where professionals do legal work, sold by the seat at a published price.Recurring software revenue
Custom deployments
Irys configured and deployed around a business's data, applications and workflows, in its own environment.Delivered on reusable software, then expanded
The shared core
Retained context, permissions, model routing and connectors. Built once, improved through product development, used by both.Why each deployment starts further ahead
Economics
Recurring revenue that grows with usage.
Software at the core
Seat subscriptions and recurring platform fees, not one-off projects.
Expands with usage
More workflows and more teams create more surface area.
Model-agnostic economics
Work is routed on quality, latency and cost.
Deployment compounds
What begins bespoke becomes reusable product infrastructure.
Market
We started with legal. The same system now runs in other industries.
Legal combines complex reasoning, private data, institutional memory, permissions and high-stakes review, which made it the hardest place to prove the system. Nothing in the architecture is specific to legal, and it now runs in banking, industrial groups and consulting.
- LegalLive
- BankingLive
- IndustrialLive
- ConsultingLive
- DefenceIn pilot
- InsuranceIn pilot
Team
Founded by a lawyer, an engineer and an AI researcher.
Former international disputes and asset finance lawyer at a global law firm, with his own book of institutional clients. Spent years inside the workflows Irys was built to change.
Leads architecture, engineering and enterprise deployment, in customer clouds and on-premises. Three-time founder: scaled Kluster to 400,000 users as CTO, then led engineering for AI agents running in 400+ Salesforce orgs at Clientell. Founding engineer at Pixxel.
Advisor: Eric Friedman, former Chairman and Executive Partner, Skadden.