Budget Management
In ANTS Platform, budgets and spend limits are part of Budget AI, the built-in AI gateway. Because Budget AI sits in front of your LLM providers, it can refuse a request that would exceed a budget before the provider is ever called.
ℹ
Budgets apply only to traffic that goes through Budget AI. Traffic you send to the platform with the SDK or OpenTelemetry is tracked and costed, but not limited. Budget AI needs the llm-gateway entitlement and a feature flag, and is turned on per project. See Budget AI for what you need.
What Budget AI budgets cover
- USD budgets per day, week or month.
- Scopes: per key, per tag, per org unit and per project.
- Either Stop requests (hard), which refuses requests once the budget is spent, or Alert only (soft, the default), which records a threshold crossing on Budgets & Limits when spend passes each alert percentage.
- Requests-per-minute and tokens-per-minute limits per key and per project.
Set them up under Budget AI → Budgets & Limits. See Budgets & Limits for the full setup and behaviour.
Watching spend without a gateway
If your traffic does not go through Budget AI, use the Gauge AI views to watch spend:
- Gauge AI → Cost Dashboard for spend over time by model, user, trace name and environment. See Cost & Latency Dashboards.
- Gauge AI → AI Optimizer to lower the cost of a specific agent. See AI Optimizer.
Next steps
- Budget AI: the gateway, keys and model routing.
- Cost Tracking: attribute spend by user, environment and tags.