# Quickstart

> What Humanload is, then five steps from your invitation to an agent that has read how you want to be asked.

Validated against app `ae3b0ef2` and Collector `0.9.0` (`54621da`), 11 November 2026.

Humanload lets you say once how you want to be asked, and the AI agents your teammates work with are told to read it
before they ask you for a decision, a review or an opinion. On a Mac, you can also install its Collector
(*Collecteur* in the French interface), the program that counts what supervising your own coding agents asks of you.

Humanload is in closed beta: you need an invitation for one e-mail address. Steps 1 and 2 need a browser. Steps 3 to 5
need a terminal.

> **Not a developer?**
>
> Do steps 1 and 2, then stop. Once your profile is saved and shared, the agents of your teammates can read it; you
> install nothing. To connect an assistant without a terminal, read
> [claude.ai and Claude Desktop](/connect.md#claudeai-and-claude-desktop): that path is not checked yet.

> **Using a coding agent?**
>
> Paste this to your agent. It reads this handbook, connects itself and checks the result; you approve the sign-in in
> your browser.
>
> ```text
> Set up Humanload for me. First read https://docs.humanload.dev/llms.txt.
> 1. Add the MCP server https://app.humanload.eu/api/humans/mcp under the name "humans" (HTTP transport, OAuth). I approve the sign-in in my browser myself.
> 2. Call get_profile with id "me" and show me the instruction lines you received. If the result is {"status":"none"}, tell me to open https://app.humanload.eu/me/start. Do not write a profile for me unless I ask; to ask, I say "Set up my Humanload profile".
> 3. Do not install anything else, and run no humanload command, unless I ask.
> ```

## 1. Sign in from your invitation

You were invited for one e-mail address. Use that address, and no other.

1. Open `https://app.humanload.eu` in your browser.
2. Enter the invited address on the sign-in page. A one-time code arrives in that mailbox.
3. Type the code.

You should see the page **How should colleagues' AI assistants write to you?**. In the menu of your personal space,
**How to ask me** comes first and **Connect your agent** next.

## 2. Say how you want to be asked

That page is at `https://app.humanload.eu/me/start`. Each of its three questions is optional, and nothing is selected
for you.

1. Answer the questions you want: which of three example messages you would rather receive, plain words or technical
   detail, and when a colleague's assistant may reach you right away.
2. Choose **See what an assistant is told**.
3. Under **What a colleague's assistant is told**, read one line per choice. Tick **Remove** next to a line you do not
   want.
4. Fill **Your name in your Teams**. A first name or a nickname is fine. A Team (*Équipe* in the French interface) is
   the group of people who share the same team view.
5. Choose **Save and share with my Teams** or **Save, only me for now**.

You should see **Saved**, with the lines an agent is told, in English, such as `Open with the decision you need from
the recipient.` [How to ask me](/how-to-ask-me.md) explains who reads them.

Once your own agent is connected (step 3), there is a second way: say "Set up my Humanload profile" and describe, in
your own words, how you want to be asked. [How to ask me](/how-to-ask-me.md#in-your-own-words) gives the details.

## 3. Connect your agent

This step is written for Claude Code, the one client seen working, against a local copy of the server.
[Connect your agent](/connect.md) says what was checked, and covers the other clients.

1. In a terminal, run:

   ```sh
   claude mcp add --transport http --scope user humans https://app.humanload.eu/api/humans/mcp
   ```

2. In Claude Code, run `/mcp`, choose `humans`, and start the sign-in.
3. Your browser opens the page **Connect an agent to Humanload**. Under **Teams it may read**, each of your Teams is
   ticked: keep the ones the agent may read, then choose **Allow**. With none ticked, the agent finds no colleague.

You should see the agent under **Connected now**, on the page **Connect your agent** of the app
(`https://app.humanload.eu/me/connect`).

## 4. Check that it read your profile

Say this to your agent:

```text
Read my Humanload profile and tell me how I want to be asked.
```

You should see the lines of step 2 in its answer. Reload the page **Connect your agent**: it says "Your agent last
read your profile on", then the date and the time, in UTC.

## 5. Optional: install the Collector

Do this step only on a Mac that runs macOS 26 or later, where you use Claude Code and Homebrew.
[Install the Collector](/install-the-collector.md) lists the conditions and what each command changes.

```sh
brew install cmdai-hq/tap/humanload
humanload setup
humanload login
humanload doctor
```

You should see, on the last line:

```text
Connection and contract compatible.
```

## Next

- [What leaves your computer](/what-leaves-your-computer.md): every field the Collector sends, and who sees what.
- [Remove everything](/remove-everything.md): uninstall, erase and disconnect, in order.
- [Troubleshooting and limits](/troubleshooting.md): what to do when a step fails, and what this beta does not do.
