RBAC & Governance
Total Control over
AI Spend.
Prevent budget overruns with atomic, sub-millisecond spending limits. Scope keys to specific models, providers, and environments.
Model Scoping
Restrict keys to specific models (e.g., 'gpt-3.5-only'). Prevent accidental usage of expensive reasoning models.
Elastic Budgets
Set strict monthly or daily spending limits. Requests are rejected in <1ms if the budget is exceeded.
Role-Based Access
Granular roles for Owners, Admins, and Developers. Control who can create keys and view spend.
# Atomic Budget Reservation (Lua)
local current = redis.call('get', key)
if current + cost > limit then
return "REJECT"
else
redis.call('incrby', key, cost)
return "ALLOW"
end
Sub-Millisecond
Enforcement.
Traditional API gateways enforce budgets asynchronously, leading to "over-drafts" during high-traffic spikes.
Hyperion uses atomic Redis Lua scripts to reserve budget for every request before it hits the provider. If a key is out of budget, the request is rejected in 0.3ms.
Zero-Overdraft Guarantee