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

# Site knowledge

> Share website goals, brand guidance, and rules with your team and agents.

Site knowledge keeps goals, audience, brand guidance, and rules in one place. It is shared across
all Branches. You can edit it in [site settings](/docs/studio/project-settings) or through the CLI.

## Read knowledge

```bash theme={null}
fimo knowledge get --json
fimo knowledge get --project <project-id> --json
```

Without `--project`, the command uses your linked site. `get` returns the Markdown `content`,
`updatedAt`, and `updatedBy`. Pretty output prints the Markdown. Empty knowledge does not prevent
work on the site.

## Save knowledge

Save the complete Markdown:

```bash theme={null}
fimo knowledge set --file knowledge.md --json
cat knowledge.md | fimo knowledge set --stdin --json
```

Each save replaces the current Markdown. An empty file clears the knowledge. Content can contain
up to 10,000 characters. `--file -` also reads from stdin.

Managers and explicitly authorized agents may edit. Asking an agent to work on the site does not
allow it to save new shared guidance automatically: ask it explicitly to update site knowledge.
Changes apply on the next agent context refresh. Switching Branches does not switch knowledge.

## Use with agents

Fimo agents receive site knowledge as context. External agents following the Fimo skills read it
before recommendations, planning, design, copy, or coding. MCP clients can search for `knowledge`
and execute `knowledge.get` and `knowledge.update`.

An agent can continue when knowledge is empty, using your request and the site. It should ask
only about information necessary for the current task. In Settings, **Draft from website** helps
prepare guidance from code, labels, and sampled CMS copy on the default Branch. It proposes tone and terminology using examples and marks inferred goals as suggestions. Code is read at a fixed revision; CMS copy is sampled when the draft runs and may include unpublished content. This is not a crawl of the live website. Review and save the draft before it becomes shared knowledge.

If you have unsaved edits, confirm before replacing them with an AI draft. You can leave Settings
and return in the same browser tab to recover a running draft, including after a reload. Temporary
drafts expire after one hour. Save applies the result; Cancel discards it.

Knowledge included in a chat follows the same chat access rules as other tool results.
Reading or updating the saved knowledge still requires the corresponding site permission.

The draft uses different evidence for each topic:

* **Design:** HTML/component structure, CSS, and theme tokens, described as observed styles.
* **Voice and terminology:** labels, headings, buttons, page copy, and CMS extracts, with examples.
* **Offerings:** product, service, and about-page content.
* **Audience and goals:** value propositions and calls to action, marked as suggestions.
* **Rules:** existing confirmed guidance; repeated website patterns do not become binding rules.

The draft is an editable Markdown brief with Brand, Target, Goals, Rules, Competitors, Tone of voice, and Design preferences sections. Unsupported sections say “Not specified.” Inferred goals or preferences are marked as suggestions; conflicts with confirmed guidance are flagged in the relevant section. Retrieval details do not appear in the brief.

## Copy or export a site

A site fork or export includes the current knowledge only when you have permission to read it.
Public website access alone does not include private guidance. Importing an archive with knowledge
saves that Markdown in the new site; older archives start with empty knowledge.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.