Keera Gateway is a Swiss LLM gateway: one endpoint for every AI request, including the providers your teams already use. You decide who may use which model, with which data and at what cost. It runs in Swiss data centres or in your own cluster.
The problem
Without clear rules, teams just get going: an account here, a key there.
Six months in, no one can still see which models see which data, or who allowed it.
No budget per team, no limit, no warning. What AI costs only shows up on the invoice.
We plan cloud exits and still hang on OpenAI, Anthropic, Microsoft and Google.
Product team
CI/CD
Data science
OpenAI
Anthropic
Azure OpenAI
Nothing in between - no policy, no budget, no log
The solution
The answer: one base URL, with your rules in front of it.
One API
A single base URL for everything, compatible with OpenAI and Anthropic. Switching provider means changing one variable.
Guardrails
Rules per team, API key and data class. Secrets and personal data stay out.
Budgets
Cost per team and cost centre, hard limits, an alert before the invoice.
Observability
Every request in the audit log and in your SIEM. Block egress, and shadow AI is over.
Routing
Small tasks to the local model, heavy reasoning to the large one.
Sovereignty
Swiss providers first, those abroad blocked until you open them. Prompts and answers are never stored or trained on.
Keera Gateway governs the models you already pay for, and those in the model catalogue. Keera Engine is the next step, when you want to run open models yourself.
The endpoint, driven from a web UI and a CLI. Your tools, agents and notebooks point at it and run as before; everyone else chats in Keera Chat. Its source is there to read on GitHub (Keera Community Licence).
The next step: an inference engine for any open model vLLM can serve, built from Kubernetes, vLLM and llm-d. On Swiss GPUs or in your own cluster.
The same gateway each time. You choose where it runs.
To start today: a shared instance with no base fee, live in an afternoon.
For an instance of your own, at a Swiss cloud provider, on a private connection.
When nothing may leave your house: in your own Kubernetes cluster, air-gapped if you want.
The story
She starts out as a bystander in a server room, follows the data into a hall with no address, and comes back with the gate she was missing there. Her adversary is an invented corporation called Omnivor.
The Omnivor corporation does not exist. The clauses and the data centres we built it from do.
Have a question or a comment? Need a quote for on-premise or an SLA? Write to us.
