Choose the models agents use

Models are chosen by Policy per task class, never in code, so changing a model changes no Interface. One Policy names a model for each class, such as draft, classify or review, with ranked fallbacks.

What you getCopy Anchor Link

• 1model Policy: one model per task class, with ranked fallbacksPolicy
• 1record of every model call, with its model, tokens and costAction

Your agent doesCopy Anchor Link

# Policy.update
curl -X PATCH https://api.sb/leaseline.app/policies/policy_models \
  -H "Authorization: Bearer $STARTUP_BUILDER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"rule":"draft: fast. review: strong, then fast."}'

import { $ } from 'startup-builder'

await $.Policy.update({ id: 'policy_models', rule: 'draft: fast. review: strong, then fast.' })

npx startup-builder do Policy.update --json '{"id":"policy_models","rule":"draft: fast. review: strong, then fast."}'

{
  "name": "do",
  "arguments": {
    "code": "return await $.Policy.update({ id: 'policy_models', rule: 'draft: fast. review: strong, then fast.' })"
  }
}

Change itCopy Anchor Link

• Use the strong model for every refund reply.
• Show me how often the first choice fails.
• Cap model spend at $200 a month.

Under the hoodCopy Anchor Link

Systems

Ledger

Nouns

Policy, Agent, Limit

Events

Policy.updated

When a model times out or refuses, the gateway tries the next in rank, and the fallback is an Event. AI models covers the gateway, cost and Limits.