Steps Reference

Steps are the atomic units of execution within a workflow.

steps.run(id, func)

The most basic step. Executes an arbitrary async function.

• Automatic Retries: If it throws, Cencori retries (default 3x) with backoff.
• Timeout: Default 60s (configurable).

const result = await cencori.steps.run('my-step', async () => {
  return await db.query('SELECT 1');
});

ai.generate(config)

A specialized step for LLM inference.

• Model Agnostic: Switch between GPT-4, Claude, Llama easily.
• Structured Output: Enforce JSON schemas (Zod).

const analysis = await cencori.ai.generate({
  model: "claude-3-5-sonnet",
  prompt: "Analyze this...",
  schema: AnalysisSchema // Optional Zod schema
});

workflows.waitFor(eventId, options)

Pauses the workflow until an external event occurs.

• Timeout: Required (e.g., '24h').
• Idempotency: Safe to call multiple times.

const event = await cencori.workflows.waitFor('approval', { timeout: '1d' });

steps.sleep(duration)

Pauses execution for a set time using setTimeout logic but persistent (survives server restarts).

await cencori.steps.sleep('1h'); // Sleep for 1 hour

Flow Control

Since Workflows are just Typescript (or Python), you use standard language features for logic.

Branching (if/else):

if (user.isVip) {
  await steps.run('vip-path', ...);
} else {
  await steps.run('standard-path', ...);
}

Loops (for/while):

for (const item of items) {
  await steps.run(`process-${item.id}`, async () => {
    await process(item);
  });
}

Error Handling

Use try/catch to handle step failures gracefully.

try {
  await cencori.steps.run('risky-step', ...);
} catch (err) {
  await cencori.steps.run('compensation-step', async () => {
    // Rollback or alert
  });
}