Get started

Begin with a portable contract or specification, then choose the integration that owns execution. llm-core deliberately has no hello-world local agent loop.

import type {
  AgentDefinition,
  AgentResult,
  AgentRunner,
  AgentStartRequest,
} from "@geekist/llm-core/agent/runtime";

declare const runner: AgentRunner;
declare const definition: AgentDefinition;
declare const request: Omit<AgentStartRequest, "agent">;

const prepared = await runner.prepare(definition);
const run = await runner.start({ ...request, agent: prepared });
const result: AgentResult = await run.result();

console.log(result.status);

The runner in this example is supplied by the host or a runtime integration. It is not provided by the kernel. A future qualified adapter may bind a native framework, but no framework becomes supported until its exact adapter subpath and conformance evidence are published. Native session, graph, checkpoint, and workspace state do not become portable merely because a runner implements AgentRunner.

For specification-driven work, load and review external intent, then compile it through an explicit adapter target. Compilation does not execute the result.

Continue with agents and runners, workflow intent, or the architecture boundary.