Attachments & artifacts
CherryPi distinguishes attachments (files you give the task) from artifacts (files the task produces). Both are stored in the agency's S3 bucket and served through the agency, so the bucket never needs to be public.
Attachments
Add files by dragging them into the console, pasting a screenshot, or using the paperclip in the composer. Limits are controlled by the agency:
| Setting | Default |
|---|---|
| Max size per file | 25 MiB (CHERRYPI_ATTACHMENT_MAX_BYTES) |
| Files per message | 10 (CHERRYPI_ATTACHMENT_MAX_PER_MESSAGE) |
When a task starts, the agent pulls the task's attachments into its workspace under attachments/. Images are also passed to Pi as image content, so the model can see them directly. Text files and documents are passed by path.
Attachments stay available in the Attachments tab of the task's details drawer.
Artifacts
An agent publishes an artifact with the bundled setTaskArtifacts tool. An artifact has:
- a name and optional description;
- one or more parts, each a file, text or structured data;
- an optional media type.
Artifacts appear in the task's Artifacts tab. File parts can be previewed or downloaded. They are the durable output of a task: markdown reports, screenshots, generated code, test logs.
Artifacts are attached to the task, not to a message, so they stay visible after the conversation scrolls.
Links
The setTaskUrls tool publishes links separately from artifacts. They appear in the Links tab and on the task list, for example a pull request or a deployed preview.
Storage operations
- Uploads are proxied by the agency with the user's session or token.
- Agents download and upload with their own personal access token.
- Deleting a task deletes its attachments and artifacts (subject to bucket lifecycle rules).
- Back up the bucket together with the database. The database stores metadata and object references, not file bodies.
Retention
Attachment and artifact retention follows the bucket's lifecycle policy and the agency's task retention. Configure versioning and an expiry policy on the bucket if you handle large outputs.