Cloud/AI Governance/FinOps Guard
FinOps Guard
The cost guard of your fleet: cap array responses, compress tool schemas, set your cost attribution rate and let the policies pay you back on every request before your budget notices.
FinOps Guard is the policy side of the money story. AI Spend shows what you spent; this screen decides how much you spend next. Its rule, in the console's own words: "Apply response and payload controls before requests reach your budget limits."
Cost Guard
Apply response and payload controls before requests reach your budget limits.
Truncates array responses after this many items
Toon Compression
Compress tool descriptions to reduce token overhead from schema definitions.
Cost attribution rate
Estimated cost per 1M tokens (USD), used to calculate spending in analytics.
These defaults are inherited by new servers. Each server can override these settings individually from its detail page.
The mockup is the real editor: drag or click the slider, flip Toon Compression, change the rate. The defaults are the console's own: 50 items, compression off, $3 per million tokens.
Max Array Items: the budget cap for fat responses
The slider controls one number with outsized impact: "Truncates array responses after this many items." An upstream API that answers a search with 400 rows becomes 50 (or whatever you set), bounded from 5 to 500 with ten tick marks. Everything past the cap never reaches the model, so it never costs you tokens, and the response stays fast. This is the mechanism behind the Cost Saved counter that AI Spend and Security Posture chart: every truncation this slider authorizes is money back.
Toon Compression: smaller schemas, same tools
One switch with a precise job: "Compress tool descriptions to reduce token overhead from schema definitions." Every connector exposes its tools with schema definitions, and those schemas ride along in context on every call. Compression shrinks that overhead while keeping the tools callable; for fleets with many connectors it is a standing discount on every single request.
Cost attribution rate: price your traffic your way
The number input sets the "Estimated cost per 1M tokens (USD), used to calculate spending in analytics." The default is $3.00; if your model's pricing differs, set yours and every dollar figure in the Reports (Estimated Cost, FinOps Savings, Cost per Request, FinOps ROI) converts to your economics. The sliders and the rate save as you adjust; no save button to forget.
Inheritance: defaults now, exceptions later
The console closes with the rule that keeps this manageable at scale: "These defaults are inherited by new servers. Each server can override these settings individually from its detail page." Set the fleet baseline once here, and override per Connector only where the numbers justify it. The footer links the console's FinOps documentation for the deep rules.