First Flow

Create hello.ts:

import { flow, llm, selectBackend } from "@twelvehart/orcats";

await flow()(async () => {
  const selected = selectBackend({
    default: "codex",
    perBackend: {
      codex: { approvalPolicy: "never" },
      opencode: { model: "openai/gpt-5.5" }
    }
  });

  try {
    const conversation = llm().autonomous(selected.backend, {
      prompt: "Inspect the repository and summarize the test command."
    });

    const outcome = await conversation.awaitResult();
    if (outcome.type !== "success") {
      throw new Error(`Backend failed with outcome: ${outcome.type}`);
    }

    console.log(outcome.result.output);
  } finally {
    await selected.shutdown?.();
  }
});

Run it:

orcats --backend codex hello.ts

selectBackend() lets the CLI select the backend through ORCA_BACKEND. ORCA_BACKEND_MODEL overrides the configured model for the selected backend.

Use terminal().emit(...) or reporter().emit(...) for concise progress during a flow. The CLI renders those facts on stderr; keep stdout for payloads that callers may pipe or parse.

Use selected.shutdown?.() in a finally because OpenCode owns a managed opencode serve process.