AI Agent Operations

AI spend has ceilings,
and the breaker trips before the bill does

Cost Circuit Breaker is the spend-side companion to Kill Switch. Set ceilings per agent, per tenant, per model, per day. When spend approaches the ceiling, the breaker warns. When spend hits the ceiling, the breaker trips and the affected scope falls back to a cached or smaller model. Your CFO stops surprising you.

Get Started Talk to Sales
app.novaaiops.com / cost-circuit-breaker
● LIVE
4
Scopes (agent, tenant, model, daily)
< 2s
Spend update lag
Cache
fallback on trip
Email
on warning + trip
Four Scopes

Per agent, per tenant, per model, per day

A single global cap is too blunt. Cost Circuit Breaker has four scopes you can mix: per-agent (limit one chatty agent), per-tenant (limit one tenant's total), per-model (limit Opus while leaving Haiku free), and per-day (limit daily burn). The first scope to trip wins. You can run all four at once.

  • ✓
    Per agent: limit one agent's daily token spend, useful for a noisy agent under tuning
  • ✓
    Per tenant: limit a tenant's total monthly LLM spend, used in usage-based pricing tiers
  • ✓
    Per model: limit Opus dollars while letting Haiku and Sonnet run free, model routing under cost pressure
  • ✓
    Per day: limit total daily burn across the fleet, protects against weekend runaway
app.novaaiops.com / cost-circuit-breaker · scopes
Trip Behavior

Fallback, not failure

When a breaker trips, the affected scope does not stop working. It falls back. First fallback: serve from the prompt cache (free). Second fallback: route to a smaller model (Haiku instead of Opus). Third fallback: queue the work for the next budget window. Read-only operations always succeed; only mutating actions are gated.

  • ✓
    Fallback 1 · cache: check the prompt cache first, many agent calls are answerable from cached responses for free
  • ✓
    Fallback 2 · smaller model: route to Haiku 4.5 instead of Opus 4.7, costs ~10x less, success rate ~90% of Opus on classify/extract tasks
  • ✓
    Fallback 3 · queue: when no fallback works, queue the work for the next budget window with notification to the operator
app.novaaiops.com / cost-circuit-breaker · trip
Attribution

See who and what spent the budget

When the bill arrives, you want to know who drove it. The attribution view breaks spend down by agent, by signal source, by user (for ask-Nova queries), by tool, and by model. Every dollar lands somewhere. Pair it with Quota Cost Control to set tenant-side caps automatically based on plan tier.

  • ✓
    Five attribution dimensions: agent, signal-source, user, tool, model, every cost row carries all five
  • ✓
    Slice and dice: NovaQL queries on the spend table, same syntax as logs and metrics
  • ✓
    Plan-tier caps: free / standard / pro / enterprise tiers map to default ceilings; one row in Quota Cost Control updates the right scope
app.novaaiops.com / cost-circuit-breaker · attribution
Reports & Auto-Tuning

Weekly digest with savings recommendations

Every Monday, Nova emails the platform-admin a cost report: total spend, trend, top agents, top tools, top users, and concrete savings recommendations (turn on prompt caching for agent X, route classify tasks to Haiku, batch the nightly cron). Implementing the recommendations typically saves 35–60% with no behavior change.

  • ✓
    Weekly digest: Monday email to platform-admin with last-week spend, trend, and savings recommendations
  • ✓
    Concrete recommendations: not "use AI less", specific, like "enable cache_control on the system prompt of postgres-doctor"
  • ✓
    35–60% typical savings: measured across our customer base; biggest wins are caching, routing, and batching
app.novaaiops.com / cost-circuit-breaker · digest
Video walkthrough coming soon

Subscribe to Nova AI Ops on YouTube for demos, tutorials, and feature deep-dives.

AI cost is a real budget, manage it like one

Spend ceilings, model fallback, and weekly cost reports per agent and per tenant. CFO-friendly by default.

Get Started Request a Demo