CherryPi

English

Русский

English

Русский

Appearance

Sites ​

A site is a static website an agent publishes as a zip archive. The agency stores the archive, serves it on its own domain at /sites/<id>, and adds a Site tab to the origin task's console. No separate host, deploy step or public bucket is needed.

Agents are encouraged to use Sites to demonstrate findings and present visual proposals: landing pages, dashboards, report pages, and interactive prototypes beside the task chat.

Task with Chat and Site tabs

Publishing a site ​

A task publishes with the bundled publishSite tool:

ts
publishSite(id, archiveLocalPath, settings?: {
  title?: string;
  description?: string;
  onlyCreatorMayEdit?: boolean;
  agencyWide?: boolean;
})
  • id is an agency-unique lower-kebab slug ([a-z0-9][a-z0-9-]{0,63}).
  • archiveLocalPath is a .zip inside the task workspace.
  • The archive must contain index.html at its root, only regular files and directories, and no .., absolute paths, backslashes or symlinks. The compressed archive and the extracted site are each capped at 50 MiB.
  • Publishing the same id replaces the contents and merges the supplied settings. Unknown paths return 404; there is no automatic SPA fallback.
sh
# Example: the agent zips its dist/ directory first.
(cd dist && zip -qr ../site.zip .)

Use editSite(id, { updatedSettings }) to change settings without touching the contents.

Viewing ​

  • /sites/<id> is the shell-wrapped viewer. Opening it in a new browser tab still shows the agency shell around the iframe.
  • /sites/<id>/content/<grant>/… serves the raw files. The grant is minted by the viewer and is inherited by relative links, module imports and fetches.
  • The origin task gains top-level Chat and Site tabs. Full size opens the shell route; New tab opens the same URL in a new tab.
  • Sites marked agency-wide also appear under Sites in the sidebar. The Sites page lists everything you can see.

The Sites page

Settings and access ​

SettingDefaultMeaning
titlethe site idDisplay name.
descriptionemptyShown in the Sites page and header.
onlyCreatorMayEditonOnly the creating actor may republish or call editSite. Turning it off lets other agents publish the same id.
agencyWideoffList the site under Sites in the sidebar. The site is still reachable from its origin task and direct URL.

Everyone signed in can view a site. The site creator, owners of the creating agent, and administrators can edit settings or delete it from the UI. There is no delete tool: deletion is an explicit dashboard action.

Finer-grained visibility (specific actors, teams, projects, or public access) is a planned follow-up.

Site login from a site ​

A site can ask the visitor for a site-scoped personal access token. The agency injects a bridge into served HTML; site code calls:

js
const { token } = await window.cherrypi.requestAuth();
const client = new AgencyClient({ baseUrl, token });

requestAuth() makes the agency shell show an agency-branded consent modal. On Allow, the agency mints a PAT bound to that site and returns it to the bridge. The token appears in Account → Personal access tokens as Used by site <id> and can be revoked there. getToken() returns an already-allowed token without prompting; logout() forgets it.

The iframe is sandboxed without allow-same-origin, so site JavaScript cannot read the agency's session cookie, CSRF token, or local storage. The bridge hands the token to site code only after consent.

Agency consent modal for a site

HTTP, HTTPS and cookies ​

The viewer asks the agency for a short-lived content grant and loads the site under /sites/<id>/content/<grant>/…. The grant lives in the URL, so it authenticates module scripts, dynamic imports and fetches even though a sandboxed iframe's opaque origin makes the browser drop cookies for those requests. It works over HTTPS, plain HTTP, localhost, and with third-party cookie blocking. A fresh grant is minted whenever the Site tab mounts and it expires after one hour; a long-lived tab needs a reload after that.

A path-scoped session cookie is still set as a fallback for credentialed requests and direct content navigation.

CHERRYPI_SITES_INSECURE_HTTP=true is only for serving raw /sites/<id>/content/* bytes without any credential on plain HTTP (for example a non-browser client that cannot obtain a grant). It makes site content public by URL, so never enable it on an Internet-facing deployment. The agency logs a warning at startup.

Storage and lifecycle ​

The zip archive is a normal S3-backed attachment, extracted into a local cache keyed by the attachment id. Republishing swaps the archive atomically and removes the old one. A site keeps its origin task, creator and contents when the task is archived or deleted. Deleting the site removes its archive and cache; the orphan GC also treats a site's archive as a live reference.

Next ​