Sample data. These figures are not a customer bill.Learn more
parezenparezen
Northwind AI
Scale · Sample · cloud
Home
  • Get startedSoon
  • Audits
  • Overview
Analyse
  • Spend
  • Cost per task
  • Agents
  • Model comparisonSoon
  • Budgets
  • AlertsSoon
  • Reports
  • PR cost checksSoon
  • AdvisorsSoon
  • Spend analysis
Optimise
  • RecommendationsSoon
  • Routing policiesSoon
  • Shadow testsSoon
  • CachingSoon
  • TrimmingSoon
  • BatchSoon
  • Waste detectorSoon
Gateway
  • SetupSoon
  • ProvidersSoon
  • ModelsSoon
  • API keysSoon
  • DeploymentsSoon
  • PlaygroundSoon
  • LogsSoon
Savings & billing
  • Savings history
  • Savings ledger
  • Bill match
  • Billing
  • Invoices
Settings
  • Data & privacySoon
  • OrganisationSoon
  • ProjectsSoon
  • TeamSoon
  • RolesSoon
  • SSOSoon
  • Audit logSoon
Plan: Scale · Sample ↗
DocsHelp

Sample data. Not a customer bill.

Northwind AI/Routing policies/Policy editor

Policy editor

Planned

Right-size, cascade, cache, batch, automation, YAML and history.

Support bot · ticket_triaged · version 3

Thresholds

The customer's bar. A lever that changes quality cannot cross it without a passed shadow test.

appTicket triagegatewayGatewayconditionserved modelmodelopenai/gpt-4.1conditioncascade offconditionJSON schema validmodelopenai/gpt-4.1-miniconditionconfidence >= 0.85modelopenai/gpt-4.1fallbackopenai/gpt-4.1fallbackanthropic/claude-haiku-4.5provideropenaiprovideranthropic
  1. app
    Ticket triage
  2. gateway
    Gateway
  3. condition
    served model
  4. model
    openai/gpt-4.1
  5. condition
    cascade off
  6. condition
    JSON schema valid
  7. model
    openai/gpt-4.1-mini
  8. condition
    confidence >= 0.85
  9. model
    openai/gpt-4.1
  10. fallback
    openai/gpt-4.1
  11. fallback
    anthropic/claude-haiku-4.5
  12. provider
    openai
  13. provider
    anthropic
Shadow testsAll tests for this routeSavingsLogs

What this policy will do

  • Ticket triage follows a cost-first policy.
  • A served response has to score at least 0.85 and finish inside 900 ms at p95.
  • Anything priced above $0.0120 a request is outside this policy.
  • Right-size is off, so Parezen keeps the model the app asked for.
  • Cascade is off, so there is no cheap-first attempt.
  • If the chosen model fails, fallbacks run in order: openai/gpt-4.1, anthropic/claude-haiku-4.5.
  • Output is capped at 280 tokens and must be structured, with reasoning effort none.
  • The route is synchronous, so it stays off batch prices.
  • Exact-match response cache is off for this route.
  • Semantic cache is Planned and is not applied.
  • Parezen only suggests. A person still chooses Enable or Ignore.

Simulated impact

12 sample requests on this route.

Baseline
$0.03
Actual
$0.03
Estimated spend
$0.03est.▲ 0.0%
Measured p95
765 ms
Escalations
0
Fallbacks
2
  • This sample has no cascade escalations. When one happens, its cost is included, not hidden.
  • Exact-match cache is off, so responses are not reused.
  • Semantic cache is Planned and is excluded.