Route an agent call to a model and effort from a policy file

At a glanceCopy Anchor Link

• Who: Agents and maintainers who pick a model per subagent call from one policy instead of hardcoding it.
• What: Resolve kind (how the agent is used for this call) times stakes (how much is at risk) and the call's needs to a provider's model and effort, with the gates of those stakes and the second review route for critical work.
• Use when: Before spawning a subagent, when the project keeps its model policy in .agents/tokenomics.md.
• Avoid when: When one session model is fine for everything, leave every cell on inherit.
• See also: Set up AI agent instructions, skills, subagents and commands in a repo

InstallCopy Anchor Link

pnpm add @zedtech/zedkit-ai

Latest: 1.0.0 (via changesets).

ExampleCopy Anchor Link

import { resolveRoute } from "@zedtech/zedkit-ai";
// `policy` is the project's tokenomics.md merged over the shipped defaults.
const route = resolveRoute(policy, {
  kind: "check",
  stakes: "standard",
  provider: "cursor",
});
if (!route.ok) throw route.error;
console.log(route.value.model, route.value.effort, route.value.gates.review);

Wraps: (none)