> ## Documentation Index
> Fetch the complete documentation index at: https://docs.usealmanac.com/llms.txt
> Use this file to discover all available pages before exploring further.

> ## Agent Instructions
> This handbook preview targets personal CLI 0.11.1. Check installed command help and the matching backend schema before using examples with another release. Examples with named people and placeholder IDs are fictional.

# Preferences and personal context

> Almanac uses what a person has said, chosen and corrected to make later work fit them.

Almanac uses what a person has said, chosen and corrected to make later work fit them. It can learn that someone values a short brief, a particular way of planning, or fewer connections on a flight. These are useful descriptions of a person in context; they do not give Almanac permission to act outside the conversation.

Suppose someone rejects two cheap itineraries because both require a long transfer. On the next trip, Almanac has a reason to show direct flights first and to explain the price difference. It has less reason to write “always flies direct.” The person may accept a connection for a much larger saving, or when there is no direct route. A useful preference records the tradeoff, the circumstances in which it appeared, and the exception that could change the recommendation.

A direct statement carries a different kind of evidence. “Keep my morning brief short” tells Almanac how to prepare that brief. One request for a short answer today may apply only to the present conversation. A pattern inferred from several choices is an interpretation that should remain open to correction. Dream, the longer-horizon review of a person's Almanac, is instructed to look for preferences in conversations and connected sources, distinguish inference from fact, and maintain subject-organized pages under `preferences/`. Those pages can hold a person's reasons, habits, constraints and exceptions. The current knowledge schema has no separate preference type; a preference page is maintained understanding written with the existing page kinds. [The wiki](/almanac/why-a-wiki) explains why that understanding is kept as an article rather than a row of observations.

This distinction changes how Almanac should use a page. If a travel preference was inferred from rejected options, it can guide a shortlist or prompt a brief explanation: “I put the nonstop options first because you disliked long transfers last time.” If the person now says that price matters more for this trip, the new instruction changes the recommendation. Almanac should preserve the earlier context where it still helps, without making the person argue with an old summary. [Communication](/almanac/communication) is where a known preference becomes a response the person can actually use.

The original conversation or provider record still matters. A saved page can tell an agent what to check, but it cannot establish that a preference is current forever. A correction, a different setting, or new evidence can change the best account. Almanac should revise the page when the person's pattern changes and keep the claim narrow enough that its evidence supports it. [Originals and evidence](/almanac/originals-and-evidence) explains how a reader can check the basis for a maintained claim.

Some instructions have a consequence beyond tailoring an answer. “Draft the reply, but don't send it” is a direct limit on that work. A preference for quick follow-ups might help Almanac prepare a concise private draft, but it does not authorize sending an email to another person. When an action would send, publish, book, disclose private information or change existing external work, Almanac follows the user's direct authorization and the active permission rules. A preference page can inform the proposed result; it cannot grant permission by being in the wiki. [Initiative and permission](/almanac/initiative-and-permission) explains the point at which preparation becomes an external action.