Skip to content
Documentation Marketplace
Cinatra Docs

Artifacts and files

Cinatra is the open source AI workspace for teams, and a lot of the value of a workspace is the stuff — the documents, drafts, images, PDFs, and references that flow into and out of agent work. In Cinatra these are called artifacts.

This page explains what an artifact is, how to bring files in through the prompt window, where artifacts live, and how to use them as inputs and outputs of agent runs.

What an artifact is

Section titled “What an artifact is”

An artifact is a self-contained deliverable — something you open, read, edit, publish, download, or attach. A PDF you uploaded, a markdown draft an agent wrote, a generated image, an external link, a published document: each is an artifact.

Artifacts are different from data objects (contacts, accounts, lists, campaign records). A data object is a record whose meaning is its relationships and status; an artifact’s meaning is its content and where it came from. For data objects, see Data and objects.

Every artifact has:

  • a stable identity that survives editing — the artifact keeps the same id as you revise it;
  • immutable versions, so each revision is preserved rather than overwritten;
  • an origin that records where it came from — an upload, an email attachment, agent-generated output, an external link, or a live generator;
  • a viewer that matches its type, so you can read it in place.

Uploading files in the prompt window

Section titled “Uploading files in the prompt window”

The fastest way to bring a file into a conversation is the prompt window itself. Open the plus (+) menu beside the message box and choose Upload files. You can select one or several files at once. The files attach to the message you are about to send.

When you send the message, the assistant or agent receives those files as attachments. What happens next depends on the file type:

  • Readable types (for example text, markdown, PDFs, and images, within size limits) are passed to the model so it can actually work with the content.
  • Other types are still attached — the model is told the file exists, its name, and that it could not read it directly. Nothing is silently dropped. You always know whether the model could see a file’s contents.

Uploaded files become artifacts with an upload origin, so anything you attach in chat is preserved and reachable later, not just consumed and forgotten.

Where artifacts live

Section titled “Where artifacts live”

Artifacts have a home in the workspace — the Artifacts area (/artifacts). From there you can browse the artifacts you have access to and open any one of them.

Opening an artifact (/artifacts/[id]) shows it in a viewer chosen for its type:

  • Markdown renders formatted, with the raw source available alongside.
  • Plain text shows as readable text.
  • PDFs open in the in-page viewer.
  • Images display inline.
  • Other types show a details card, with an external-open button when the artifact points at a connector-owned resource.

Every artifact detail view has a Download button so you can pull the file out when you need the original.

Because artifacts ride on the same ownership and scope model as everything else in Cinatra, an artifact made inside a Project stays with that Project’s work. See Projects and ownership.

Artifacts as agent inputs and outputs

Section titled “Artifacts as agent inputs and outputs”

Artifacts move through agent work in two directions.

As inputs (context). When an agent run needs grounding material — a brief, a source document, a prior draft — it consumes artifacts as context. The reference is pinned to a specific version at the moment it is selected, so a run always works from the exact content it was given, even if the artifact is edited later. Some agents declare what kinds of artifact each input slot accepts, and the workspace resolves the right artifact for the slot when the run starts.

As outputs. When an agent produces a deliverable — a written draft, a generated image, a compiled document — it is written back as an artifact with an agent_generated origin. It then behaves like any other artifact: you can open it, read it, download it, edit it (creating a new version), and attach it to a later conversation.

This is what lets work compound. An agent’s output becomes the next agent’s input; a file you uploaded this morning grounds a run this afternoon; a draft you edited by hand is the version the next step builds on.

Sharing and permissions

Section titled “Sharing and permissions”

Artifacts follow the same access rules as the rest of your work, so there is no separate “artifact sharing” system to learn:

  • An artifact is owned at one of the ownership tiers (user, team, organization, or workspace) — see Projects and ownership for what those tiers mean.
  • An artifact created inside a Project is scoped to that Project, and the people with access to the Project can reach it.
  • Widening who can see an artifact is the same act as widening any owned resource — raise its tier or work inside a shared Project. If you need an artifact made visible more widely than your role allows, ask an admin.

Referenced artifacts are never hard-deleted out from under work that depends on them — a version that a run or message points at is retained, so past work stays reproducible.

Section titled “Related”