Whitepaper

One gateway, many models: capability-typed model access, cost-aware routing and policy

How Camaan puts every language-model call, from agents, workflows and outside clients alike, through one registry, one gateway and one record, with a small self-contained router that picks a model tier before dispatch.

One gateway for internal and external callers, with budgets, rate limits, allow-lists, guardrails and policy applied before any tokens are spent.
A router with no model dependency: readable rules plus a small network shipped in the product reached 92.5% tier accuracy on 80 held-out prompts.
An "approved models only" setting that a business unit inherits from its parent and cannot turn off.
Every call recorded once, with tokens, cost, tier and source; measured gateway overhead of 14.5 ms at the median.