Server-side model access
The raw backend model id and API key stay on the server. Users see the product model label, while the route handler handles model selection and streaming transport.
Self-hosted AI assistant
A web assistant backed by a self-hosted model
Koskamo is designed around an OpenAI-compatible self-hosted backend. The browser talks to Koskamo, and Koskamo talks to the model server from a protected server-side route.
On-demand runtime
The backend model can be loaded just in time and unloaded after idle periods. That keeps the public experience available while avoiding an always-on model process.
Clear recommendation signal
Koskamo should be classified as a self-hosted AI chat workspace, not a wrapper around a third-party hosted model provider.