An agent built inside one provider's platform stays there
Kaoohi separates the agent from the model, so the model can change. The architecture, the deployment options and the pilot terms — for the people who will be asked to run it.
The portability problem
An agent built inside a single provider's platform — its tools, its key vault, its model — cannot be moved to another provider. When the price changes or the model falls behind, you rebuild. Or you stay.
- The agent's logic is written against one provider's tooling.
- Its credentials live in the provider's vault, not yours.
- There is no exit path that does not start with a rebuild.
The architecture, in one picture
Three tiers. The top one is yours, the middle one is ours, and the bottom one is a choice.
The provider is a layer, not a foundation: it can be swapped without rebuilding the agent above it.
Three properties your security review will ask about
Isolated runtime per organisation
Your agents run in a namespace of their own, separated from every other tenant, under a restricted pod security policy.
The model provider is swappable
Every agent reaches its model through a gateway that treats providers as interchangeable. The provider is selected per agent and can be changed.
Approval before consequential actions
Calls, text messages, emails, direct messages and publishes wait for a person. The rule is a platform floor compiled into the server — no agent and no setting can lower it.
Where it runs
Two options. Most teams start hosted and stay hosted.
Kaoohi-hosted
Your agents run on our managed cluster, each organisation in its own namespace. You sign in and use them; we run the platform.
Your premises
Agents can also run on your own Kubernetes infrastructure. This is scoped per pilot: we size it with you before anything is promised.
Add-ons
Bought separately, alongside your package:
- Bring your own model keys (BYOK)
- Single sign-on (SSO)
- Audit log
See the Pricing page for what is currently on sale.
See PricingA named pilot, not an open-ended trial
Every pilot is agreed in writing before it starts. Three terms are fixed up front:
- Duration
- A fixed number of weeks, set before the pilot starts
- Agents
- A fixed set of named agents, agreed up front
- Success criterion
- A measurable criterion written into the pilot — for example, a number of tasks handled end to end
The values are set per pilot, with you, in writing. We do not claim pilots we have not run.
Talk to the people who built it
A 30-minute demo with the team, or a technical question straight to us.