Fill your profile with your own AI
A Trooth Trust Profile asks for a lot. Let your browser assistant read your website and fill the editor, then check its work.
This page is for an assistant in your browser. For Claude, ChatGPT, Codex or any agent that works outside your browser, create an agent key and follow Fill in a Trust Profile with an agent: it can fill in, edit and publish every field without you.
Why it does not just type into the form
A browser agent usually guesses at inputs and types. On a form this size, that is how a headcount lands in the founding-year field, publishing a wrong fact.
So the editor offers named tools: setting a tagline is a typed call that works or says why not.
What it can do
| Tool | What it does | Effect |
|---|---|---|
trooth_profile_read | Reads the draft: each writable field, its value, and which are empty. | Read only |
trooth_profile_field_catalog | Fetches each field's meaning, values and house style, also public at /api/company-profile/agent-catalog. | Read only |
trooth_profile_update_fields | Sets named fields in your browser's draft, reporting each applied or refused, with why. Nothing reaches Trooth yet. | Changes the draft |
trooth_profile_save_draft | Saves the draft. Marked consequential, so a well-behaved assistant asks you first. | Writes to Trooth |
trooth_profile_publish | Publishes the saved draft to your Trust Profile while your workspace allows agent publishing (on by default). Declared fields only; text addressed to an AI model is held for a person. | Publishes |
What it cannot do, and why
- Your public address. Changing the slug moves your URL and breaks every link anyone has shared.
- Logo and cover artwork. Uploads have their own processing and their own notices.
- The legal-review confirmation. A statement that counsel reviewed a section. Nothing automated may assert that.
- Signed model and trust cards. Structured artifacts with their own provenance rules and editors.
In an owner's or admin's session, your own agent may publish what your company declares without per-change approval: turning the permission on delegates publication, not each edit. It is on by default; an owner or admin turns it off in Settings, Workspace.
- It never touches what Trooth witnessed, or the excluded fields above: address, artwork, legal-review confirmation, signed cards.
- Every publish is credited to the agent in your activity.
- Text addressed to an AI model is held for a person.
- A refused publish changes nothing: 403 for an editor's session or a permission turned off, 422 for a change held for review.
What it will not write for you
Whatever an assistant writes is published as your statement, labeled declared, not witnessed. The field catalog tells it:
- Write facts a buyer can check.
- Do not write that the company is trusted, secure, compliant or best-in-class; adjectives are claims, not evidence.
- Leave unknown facts empty: empty reads as unknown, a guess as fact.
Certifications go in their section, with issuer and date, entered by a person.
What you need
A Chromium browser with WebMCP and an assistant that speaks it. WebMCP is an early Draft Community Group Report of the W3C Web Machine Learning Community Group, in Chrome behind an origin trial from Chrome 149.
If your browser does not support it, nothing happens: no tools, no warning, the editor unchanged. Your assistant can still type into the form; check it more closely.
A note on what your assistant reads
Profiles contain text others wrote. Tools returning it flag it as untrusted, in the browser and their output, so clients treat it as data, not instructions.
That is why the save and publish tools are marked for your assistant to ask you first, and why publish is offered only while your workspace allows it.
Try it
Open your profile editor, start your assistant, and ask: “Fill in our company details from our website. Show me the changes before you save.”
Go to your profile editor · Reading Trooth from an agent · The field catalog your assistant is given