# How to ask me

> What a profile is, who reads it, what an agent does with it, the two ways to fill it, and how to change or delete it.

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

Your profile tells the AI agents of your teammates how to write to you, how to put a decision to you and when to
interrupt you. You say only what you want said: every preference is optional.

## What a profile is

A profile is one file in the HUMANS.md format, version 0.3. It holds closed values, picked from lists: there is no
free text for an agent to read. It has five parts:

| Part | What it says, with one example |
|---|---|
| `reading` | the form of a message to you: `length: short` |
| `deciding` | how to put a decision or a question to you: `asks_per_message: one` |
| `interrupting` | what is worth stopping you for, and your quiet hours: `interrupt_for: blocking_only` |
| `context` | your role, what you know well or little, the kinds of decisions you own: `decides: [release-timing]` |
| `avoid` | what to leave out of a message: `filler_intro` |

It also holds `languages`, the languages to write to you in, and `visibility`, who reads it.
[Profile fields](/reference/profile-fields.md) lists every value and the sentence it produces.

A profile holds no Count (*Compte* in the French interface: a number the Collector counts). It is about the form of a
message to you, and the service tells agents never to use it to evaluate, compare or assign work to people.

## Who reads your profile

| Reader | Reads what |
|---|---|
| You, in the app | everything you saved |
| Your own agent, once connected | your profile, even private |
| The agents of the members of your Teams | your profile, only when you shared it with your Teams |
| Anyone else | nothing |

Three more facts:

- **Your profile carries no name.** The name of your account helps an agent find you among the members of a Team:
  a lookup by name (`resolve_person`) returns it with your id. The profile an agent receives carries no name.
- **Free notes are never served.** A file an agent saved may hold a `notes` field; the service keeps it and gives it
  to no agent, yours included. Your data export contains it.
- **Each read is logged for 30 days.** The app shows you how many agents read your profile, by week. A reader appears
  as a code, never as a name.

The operator of this beta can technically read stored rows. [What leaves your computer](/what-leaves-your-computer.md#who-can-see-what-on-the-service)
states what that covers.

## What an agent does with it

The service gives a connected agent these instructions, in this order:

1. Before asking a colleague for a decision, a review or an opinion, find the person with `resolve_person`.
2. Read their profile with `get_profile`, before drafting the request.
3. Hold the request, and tell the person the agent works for, when the request is of a kind the colleague does not
   decide, or when the profile says not to interrupt now. The agent never picks another recipient itself.
4. Otherwise, shape the request as the instruction sentences of the profile say.

A profile changes the form and the timing of a message. It grants no permission, names no recipient and gives no task.

An agent is told these rules; nothing forces it to follow them. The service gives the same text to every client:
[MCP tools](/reference/mcp-tools.md#instructions-the-server-gives-agents) shows it in full.

## Fill it in the app

Do this when you have no profile yet. The page is **How to ask me** in the menu, at
`https://app.humanload.eu/me/start`; it is also where you land after your first sign-in. Once you have a profile, the
same address leads to the full form, and the quick way never overwrites a profile.

The first screen is titled **How should colleagues' AI assistants write to you?**. Each question is optional, and
nothing is selected for you: what you leave out is not stored.

1. Pick the message you would rather receive, among three ways to write the same request:

   | Card | What it sets |
   |---|---|
   | The decision first, in five lines | `reading.lead_with: decision_needed`, `reading.length: minimal` |
   | A short summary, then the details | `reading.lead_with: tldr`, `reading.length: standard` |
   | The full context, then the question | `reading.lead_with: context`, `reading.length: thorough` |

2. Answer **Plain words or technical detail?**.
3. Answer **When may a colleague's assistant reach you right away?**.
4. Choose **See what an assistant is told**. **Not now** leaves without saving.
5. On the second screen, **What a colleague's assistant is told**, read one line per choice, in English. Tick
   **Remove** to leave a line out. **Change my answers** goes back.
6. Fill **Your name in your Teams**.
7. Choose **Save and share with my Teams** or **Save, only me for now**.

You should see **Saved**, then two links: **Connect your agent** and **Fine-tune**.

**Fine-tune** opens the full form, at `https://app.humanload.eu/me/profile`. That page lists every field of the
format in four sections, and shows the exact text a teammate's agent receives.

## Fill it with your own agent

Do this when you prefer to describe yourself in your own words, or when you already keep a HUMANS.md file. Either
way, connect your agent first and keep **Save or delete my profile (How to ask me)** ticked on the consent page
([Connect your agent](/connect.md#what-the-consent-page-asks)).

### In your own words

Say "Set up my Humanload profile", then describe how you want to be asked. For example:

```text
Set up my Humanload profile: I'm visual. I have little time, and I am asked for a decision on something every week.
```

The tool's description tells your agent what to do with such a sentence:

1. Turn it into settings about the messages you receive: what to lead with, the length, visuals, questions per
   message, when to interrupt you, what you decide. It sets only what your words support.
2. Ask what name you go by, and whether your Teams read the profile or only you.
3. Show you every setting as a plain sentence, with one short example request written that way, before it saves.
4. Save exactly what you agreed to. If the service refuses the file and the agent has to change a setting, it shows
   you the change and asks again.

You should see the settings in plain sentences, then the tool answer `saved`.

What the agent sends: the resulting settings. The agent is told to keep your words to itself, to send only the
settings, to leave free notes out, and never to write a health condition, a diagnosis or a personality label into the
file. You may name one as a way in; the settings it leads to are what is saved. The service cannot
check what an agent was told: read the settings before you agree.

### With a file

1. Tell your agent what to save, and who reads it:

   ```text
   Save this as my Humanload profile with put_my_profile, private for now:
   lead with the decision needed, keep messages short, one question per message, write to me in English.
   ```

2. The agent sends a file like this one:

   ```markdown
   ---
   humans: "0.3"
   id: me
   name: Maya Example
   updated: "2026-11-11"
   visibility: private
   languages: [en]
   reading:
     lead_with: decision_needed
     length: short
   deciding:
     asks_per_message: one
   ---
   ```

You should see the tool answer `saved`. Two other answers exist. `invalid` means the format refused the file: the
answer gives one generic issue per refused top-level field, or `$` when the format does not say which, and lists every
field with its allowed values so that the agent can correct the file. A file whose frontmatter is too large gets a
single `$` issue and no list. `unavailable` means the service cannot store the file at the moment. Text below the second `---` line is
dropped.

## Choose who reads it

`visibility` takes one of three values:

| Value | In the app | Who reads the profile on this service |
|---|---|---|
| `org` | My Teams | you, your own agent, and the members of each of your Teams, now and later |
| `private` | Only me and my own agent | you and your own agent |
| `anyone_with_file` | Anyone who has the file | the same readers as `org`: this service never serves a profile outside your Teams |

### What a missing `visibility` means

A file without `visibility` is shared with your Teams. The format reads a missing value as `org`.

- In the app, you cannot save without choosing: nothing is shared by omission.
- With an agent, the tool's description tells the agent to ask you which you want. State it yourself, as in the
  example above.
- When a saved profile has no `visibility`, the app says so on the profile page and asks for the choice at your next
  save.

## Change it

Save again, in the app or with your agent. A save replaces the whole profile: nothing of the old one is kept, and the
agents of your teammates read the new one from their next request.

With an agent, send the full file every time. `get_profile` does not return everything a file holds: it leaves out
`humans`, `name`, `visibility`, `aliases`, `notes` and `interrupting.when_focused`. An agent that rebuilds your file
from what it read, then saves it, erases those fields and, without `visibility`, shares the profile with your Teams.

Tell your agent one of two things:

- "Ask me for my full HUMANS.md file before you change it."
- "Here is everything to save", followed by every preference, your name and who reads it.

Two fields are required and never served to a reader:

- `id`: 2 to 64 lower-case letters, digits or hyphens. Write `me`, as the app does.
- `name`: your name, 1 to 80 characters. A file without it is refused. The service does not use it to find you:
  lookups use the name of your account.

## Delete it

1. Open `https://app.humanload.eu/me/profile`.
2. Under **Delete my profile**, choose **Delete…**, then confirm.

You should see "Your profile and its read log are deleted." From then on, an agent that asks for your profile
receives `{"status":"none"}`.

An agent that you allowed to write can do the same with `delete_my_profile`, only when you ask it to.
