You: your record

You (in the sidebar under Identity & Vision, and under the Library's masthead) is the page that holds what HeroMind, and every AI you connect, knows about you. It is the one place in the product where a statement carries a status and needs your yes before anyone trusts it.

What the record holds

Every item is in your words, never paraphrased, and carries where it came from. Compiled pages elsewhere in HeroMind (Foundation, Life Blueprint, your portfolio documents) feed the record too, as proposals you accept.

Status

Each item is current, proposed, set aside, superseded, or expired.

• Current items are what agents read.
• Proposed items wait on the Desk. Nothing enters the record without your yes, and a proposal that sits for 14 days expires.
• Working items are dated things you said that are true for now: a project, a sprint, a question still open. Assistants may read them with the date and "not confirmed"; they never enter your compiled page and they expire on their own after about a month unless you keep one.
• Set aside items are proposals the curator judged not worth your attention: a conversational fragment, a status update, a duplicate. They are never accepted and never deleted on their own; What the curator set aside on the Desk lists them with Restore and Delete.
• Superseded items are kept as history when you change your mind. Accepting a proposal that replaces an older item retires the older one; you can also supersede by hand. The lineage of any idea, what you thought, when it changed, and why, is one tap away.
• Corrections are guard facts. When a page or an agent states something wrong about you, correct it in your words; the correction is linked to the subject and the page recompiles.

Where proposals come from

• Agents you connect, when they believe you stated a decision or preference. Every agent write to the record is a proposal; an agent cannot make an entry current, replace one, or correct one without your yes.
• The nightly sweep of your own AI coding sessions, which extracts judgment from transcripts.
• Ask conversations in the Library: the ones you keep are distilled at once, the rest overnight.
• Your assistant, when a conversation in Talk it through lands on something durable.
• Meeting notes, where the librarian finds decisions.

The Desk shows them together, in your words, with the source. Anything you dismiss is remembered as "not now", not re-proposed tomorrow.

The curator

Proposals arrive faster than anyone wants to read them, so the librarian reads them first. It never accepts anything; it decides what deserves a decision from you.

For each proposal it asks four things: who said it (your words with a quote, or an assistant's inference), what it is (a commitment, a possibility, a temporary instruction, an open question, chatter), whether it materially changes what your record already says, and where it applies and when it should stop counting. Then:

• A change to your record becomes a card on the Desk: the current wording it replaces or refines (if any), the proposed wording, why it changes the record, and the words it rests on. Expect three to five in a normal review, never a pile.
• A situational statement or an open question goes straight into working context, dated, without asking you.
• Chatter, a duplicate of what the record already says, or an assistant's reading with no quote behind it, is set aside. Set-aside items stay viewable on the Desk under "What the curator set aside", with Restore.

Once, for the entries that were written straight into your record before proposals became the rule, Review my approved record for chatter at the bottom of the Desk asks the curator to read what is already approved. It answers with one card, "Retire these N?", every entry ticked with a reason. Untick what should stay and confirm; retired entries keep their history.

It runs every night over whatever arrived since the last pass, and on demand: Curate N now on the Desk when new proposals are waiting, Curate again to re-read everything still proposed. On each card you can Accept, keep it Privately, say Not now (it parks in working context for a month), or Dismiss.

Where the words come from

Every statement shows who said it. Your own words appear under it as a quote with the date ("You said, Sep 28: …") and the quote stays even if the conversation it came from has expired. An entry an assistant inferred says so: "An assistant's reading, not your words." One filed by an agent without a quote says that too. The one-line summary on a conversation is written by the librarian and labelled "summary"; it is never treated as your position.

The tabs

• Home. A readable account in six questions: your situation, what matters to you, where you are going, how you work and want to be supported, what you currently think, what has your attention. Your own statements are quoted verbatim, and every footnote opens the record it cites. Below it, What has your attention is computed from your interests, active quests, open questions and what you have been reading lately; it is not a list you maintain. Needs you counts the changes waiting on the Desk and the working items that will expire. Still true? asks about anything you set a reminder on. Something wrong? files a correction in your words and the page recompiles.
• Record. Your statements, grouped by the question they answer, with an Approved / Working context filter and the quote under each. Retire one when it no longer holds; the history stays. Beneath them, the other layers this page is compiled from, with their own editors, and the ownership rule: Foundation owns vision and strategy per life category, Identity Center owns purpose, values, beliefs you are examining and how you describe yourself, Portfolio owns prose you wrote, the compiled page owns nothing and cites everything.
• History. Timeline, Reflections, Conversations and Changes behind one row of chips. What happened, what the librarian wrote back, what you asked, what moved.
• Ask. Ask your own record a question, with citations and a "Context used" line showing which of your statements and conversations the answer drew on. Each entry is marked by how it got there: quoted from your own words, or inferred by an agent without a quote; the librarian trusts the quoted ones more when two disagree. Ask about a time ("what did I work on this week?") and the answer lists everything from that window — your conversations and what the librarian ran for you — rather than guessing from similarity.

What agents get

An agent that calls get_context with a question and a budget receives a pack: a header on who you are, your standing preferences, the decisions relevant to what it is about to do, what happened recently, and today's plate. Items are ranked by meaning and by keyword, returned verbatim with provenance, and the whole pack fits the budget the agent asked for. It is deterministic and takes under a second, so "why did the agent think that" is always answerable by looking at the pack.

What agents do not get: proposed items, superseded items, and anything you have marked private.

Every statement an agent receives carries its provenance: your quoted words and their date, "their own words" when you stated it yourself, or "inferred by ChatGPT, no quote" when an agent read it into you. Agents are told to weigh inferred entries lower. If an agent seems to know something you never said, the label shows which agent filed it, and the Desk is where you retire it.

What you get

The same page, plus the history. You can see exactly what any connected agent sees, correct a sentence, retire a belief that no longer holds, and export the whole record as Markdown.

Rules worth knowing

• Agents propose to the record. They never write to it, and neither does the curator or your assistant.
• Your words are kept verbatim. The librarian compiles around them; it does not rewrite them.
• Nothing is deleted when you change your mind. It is superseded, and the history stays.
• The record is about you. What your agents know about the world lives in the Library.

Updated 2026-10-11 · All chapters · API reference