IRIS8 Talk to us

For the CTO

How does it fit what we run?

Your model sits behind an OpenAI-compatible API in your own environment, and IRIS8 operates it.

Your brief

What you answer for

  1. 01An architecture that does not depend on one model vendor
  2. 02Releases you can pin and roll back
  3. 03No new platform for your engineers to run

Questions

Five questions CTOs ask us

The answers we give on the first call.

  1. 01

    How does it connect to our applications?

    Through an OpenAI-compatible API. An application that already calls a model changes its base URL and keeps its code. The frameworks named in our documentation connect the same way: LangGraph, CrewAI, AutoGen and MCP servers. Your teams do not learn a new SDK.

  2. 02

    Where does it run, and what do you need from us?

    In your VPC or on-prem. We need a cloud project or account we can deploy into, GPU capacity sized to the model and traffic, and one infrastructure contact. We size the hardware in the first week of the pilot, before you commit to buying or reserving any.

  3. 03

    Which models do you use, and are we locked in?

    We start from licensed open-weight models, shortlist several for the workflow and test them against your benchmark before training. You own the fine-tuned weights. Frontier models stay reachable through your gateway, so there is no lock-in to one model vendor.

  4. 04

    How do releases and rollback work?

    Each version is pinned. A new version is released in stages, and only after it passes your benchmark and you approve it. The previous version is kept, so rollback is a decision your team makes, with nothing to rebuild. The release calendar is yours.

  5. 05

    What can my engineers see?

    The same console our engineers work in: candidate models, training runs, benchmark results, live gateway requests and the model record. Request logs stay in your environment, so your own monitoring can read them. Nothing about the model is visible to us and hidden from you.

The stack

Your apps and frameworks
  • LangGraph
  • CrewAI
  • AutoGen
  • MCP servers
EndpointOpenAI-compatible APIChange the base URL
Models
  • Open-weight base model
  • Your fine-tuned model
  • Frontier model, optional
Runs on
  • Your VPC
  • Your data centre

One request, through an endpoint you already know how to call.

Next step

Bring one workflow to a 30-minute call