GUIDES

Engine

A model can answer; the Engine can finish the work. It’s the execution layer that wraps adaptive intelligence in deterministic structure, so outcomes are repeatable, verified and shippable.

Workflows

A workflow is an ordered set of steps. Each step can call a model, run a tool, generate media, or branch on a result, giving you control around the parts that should be non-negotiable.

import { Taskclan, workflow, step } from "@taskclan/sdk";

const publishPost = workflow({
  name: "publish_post",
  steps: [
    step.run({ id: "draft", profile: "t1-flow", goal: "Draft a post from these notes" }),
    step.generate({ id: "hero", modality: "image", promptFrom: "draft" }),
    step.evaluate({ id: "check", policy: ["brand", "rights"] }),
    step.tool({ id: "publish", tool: publishToCms, when: "check.passed" }),
  ],
});

Run a workflow

const taskclan = new Taskclan({ apiKey: process.env.TASKCLAN_API_KEY });

const run = await taskclan.engine.run(publishPost, {
  input: { notes: "Launch recap: 3 highlights, friendly tone" },
});

console.log(run.status, run.steps.publish.output);

The evaluation layer

Before anything ships, the Engine can verify output quality, policy compliance, rights and business outcome. Gate side-effecting steps on evaluation so nothing risky goes out the door.

Tool execution

Steps can act through the same typed tools your agents use, APIs, browsers, internal systems, so a workflow doesn’t just produce content, it delivers it.