Capability
AI models
The language models we build with, hosted or on your own hardware.
Why us on this
We are not tied to one model vendor. A model is picked per task from a shortlist, by running it against the client's own documents and a fixed evaluation set: accuracy, cost, latency, and where the data is allowed to go. Hosted APIs where that is acceptable; open-weight models served on the client's own hardware where it is not. Models change every few months, so they sit behind a thin layer that makes swapping one a configuration change, not a rebuild.
Products in this capability
What we actually reach for, and what each one is good at.
OpenAI GPTOpenAI
OpenAI's large language models, used through the API.
Runs open-weight language models on your own hardware.
GemmaGoogle
Google's open-weight language models, run on-premise.
High-throughput serving for open-weight models.
Anthropic's Claude models, used through the API and in our own engineering.
Google GeminiGoogle
Google's multimodal models, strong on long documents and images.
Alibaba's open-weight model family, from small to very large.
xAI's Grok models, available through an API.
Capable open-weight general and reasoning models.
Open-weight models with very long context windows.
Containerised inference microservices for running models on NVIDIA GPUs.
Azure OpenAIPart of Microsoft Azure
OpenAI models inside the client's own Azure tenant.
Talk to us
Thinking about AI models?
Tell us where the work stops today. We will say plainly which tools fit it, what moving it takes, and when this is the wrong approach for what you are trying to do.