> ## 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.

# What the agent can see

> Almanac can work with the person's knowledge, connected originals, files and a desktop interface.

Almanac can work with the person's knowledge, connected originals, files and a desktop interface. It does not automatically see everything the person sees. Before referring to “this email,” “the file on the right,” or “the page you have open,” establish which object the current message or an available tool actually identifies.

The agent receives the conversation, its enabled tools, results of its own calls, and a small runtime snapshot. That snapshot may say whether personal Almanac is configured, which accounts were recently connected, the session surface, and the session or task ID. It does not include the active desktop tab, the user's foreground browser page, or a live screen feed. An account snapshot is orientation; check the relevant account and object when access or current state matters.

## A result, a card, and an original are different

An `almanac` command returns output to the agent. The desktop may also turn a direct command into a compact activity row for the person. That row is a rendering of the tool call, not another source the agent has read. Long terminal output can be truncated, and a search or list may have another page. Narrow the query, follow a cursor, or read saved full output before claiming coverage.

`show_widget` returns a reference that the desktop can render as an email, calendar, or connection card. A person can click an email or calendar card to open its corresponding reader. A connection card opens the Connections or add-provider pane; it does not itself connect the account. Successful presentation does not mean the agent read every email message, checked calendar availability, opened the viewer, or knows whether the person clicked. Read the original through the appropriate account tool for any claim that depends on its contents. A chat citation similarly gives a reader a route to a source; it does not repair an incomplete read.

An HTML or SVG answer may become an artifact card automatically. That card is a version of the answer in the interface, not proof that a local file was written or opened. A file created by a tool exists in that tool's execution environment. On a remote gateway, its path may not be a path on the person's computer. If the session exposes `open_preview`, use it when a Files pane is the useful handoff; do not report that a pane opened merely because a path was printed.

## Check the view that is actually available

In a desktop session, `read_preview` can inspect the active **Files** preview, and `read_window_below` can give window metadata. Neither gives a general view of the user's screen or browser tab. Browser tools inspect the browser session they control when called. A user may also explicitly attach an app snapshot; treat its image and accessibility text as a dated observation, and seek a fresh view if later actions could have changed it. Tool availability varies by session: a background or chat-connected agent should not assume it can focus a desktop pane.

When uncertainty changes the answer or action, say what you did observe and what remains unseen. “I placed the thread card beside the answer” is a useful result. “I opened and checked it for you” requires a separate observed action. Match the handoff to the person: give the finding first, then the card, citation, or file that lets them inspect the relevant original. For exact command syntax, use current help; for evidence and source identity, follow [Originals and evidence](/almanac/originals-and-evidence).