All guides

10 min read

Run your first agent

Describe a task in plain English, give the agent a tool to reach for, and watch it work. Ten minutes from an empty workspace to a run you can read end to end.

  1. 01

    Connect one integration

    Open Settings → Integrations and authorize a single service — Slack and GitHub are the easiest first connections. Helios stores the credential envelope-encrypted and attaches it server-side, so it never appears in a prompt or in the agent's sandbox.

  2. 02

    Create a workflow

    A workflow is an agent with a standing task. Give it a name that says what it does — `Daily standup digest` beats `Workflow 1` when you have twenty of them and a run fails at 3am.

  3. 03

    Write the prompt in plain English

    Say what the agent should do, which sources it should read, and what a good result looks like. Be specific about the boundaries: what to skip, when to escalate, and what to do when it is unsure. The prompt is the whole configuration — there is nothing else to wire up.

  4. 04

    Give it the context it needs

    Every agent runs on Claude Sonnet, tuned for long-running tool use — so the configuration that matters is context. If the task works in a codebase, pin the GitHub repository; if it needs a threshold or an API key, store it as a variable or secret.

  5. 05

    Run it manually and read the run

    Trigger the run yourself the first time. The run record shows every tool call the agent made, the arguments it passed, and what came back. This is where you learn whether your prompt said what you meant, so read it before you automate anything.

  6. 06

    Refine the prompt, then add a trigger

    Adjust the sentence that produced the wrong behavior and run it again. Once two or three manual runs look right, attach a trigger and the agent starts picking up the task on its own.

Try it in your workspace.

Start free, connect a tool, and describe the first task in plain English.