Artifacts

When your AI does a piece of work that ends in a thing — a document, a comparison chart, a one-pager, a meeting brief, a contract draft, an exported CSV — that thing is an artifact.

The word matters, because “file” is already taken. Files are inputs you upload. Artifacts are outputs your AI produces. Keeping them named differently keeps it clear who made what.

What an artifact looks like

An artifact has:

A title. Short, specific, picked by your AI or by you.
A body. Usually Markdown, sometimes a table, sometimes a structured document.
A type. “Brief”, “comparison”, “spec”, “summary”, “transcript”, “checklist”, “draft email”, and so on. The type is mostly for organizing.
An owner. The entity it was produced for: a project, an idea, a contact, a goal, or you directly.
A version history. Every revision is kept. You can go back.
A status. draft while it’s a work in progress, approved when you’ve signed off, published once it’s been sent or used somewhere.

You can view an artifact in the web app, export it, copy it to your clipboard, email it to someone, or use it as the input for the next thing your AI does.

How to ask for one

Most artifacts come from asking. The pattern is “make me X about Y.”

“Make me a one-pager on the redesign project for the team meeting tomorrow.”
“Draft a press release announcing the v2 launch.”
“Compare the three contractor quotes side by side.”
“Summarize the last month of activity on the Acme account.”
“Spec out the database schema for the new feature.”

Your AI produces the artifact, names it, files it under the right owner, and shows it to you. If the owner is ambiguous (“which project?”), it asks first.

The same shape works for short outputs you might not even think of as “documents”:

“Draft a reply to Mike accepting the quote.”
“Give me a checklist for tomorrow’s beach trip.”

These produce small artifacts. Some of them you’ll use once and forget. Some of them you’ll come back to. Either way they’re saved, in case you want them later.

Asking for a revision

Artifacts are rarely right the first time. The revision pattern:

You: Make the tone more direct, drop the second paragraph, and add a section on pricing.

Your AI: [produces a revised version of the artifact, keeping the previous version in history]

You can also point at a specific section (“rework the introduction to start with the cost”) or do larger structural moves (“turn this into a slide deck outline instead of a memo”).

Each revision is a new version. The artifact’s history pane shows them in order. You can flip back to an earlier version, or compare two versions, or fork an artifact and let two versions exist side by side.

Approval, publish, and using artifacts

A few things you can do with an artifact once it’s good:

Approve. Marks it as the version that’s ready. Future references default to the approved version.
Export. Markdown, PDF, plain text, or copy-to-clipboard.
Send. If you’ve connected email or another channel, you can have your AI send the artifact (as an attachment, or pasted into the body) to a specific person or contact.
Reuse. Tell your AI “use this brief as the basis for the kickoff email” and it does, keeping the source artifact linked.

Where artifacts live

Every artifact has exactly one owner. The owner determines where the artifact shows up:

A project’s Artifacts tab shows artifacts owned by that project.
A contact’s Artifacts section shows things you’ve written about that person (briefings, profiles, conversation notes).
An idea’s expanded view shows the lenses (comparison charts, risk analyses) it has produced.
Your Artifacts view in the sidebar shows everything across the instance.

Cross-references work even when the owner is fixed. An artifact owned by the kitchen project can be linked from the contractor’s contact page. The artifact itself doesn’t move; it just gets a reference from the contact.

Artifacts your AI produces on its own

Not every artifact comes from you asking. Some get produced as side effects of automation:

A goal that runs every Monday morning and writes a weekly review produces a fresh artifact each Monday. They accumulate in the parent project’s Artifacts tab.
A playbook that fires when you click “comparison chart” on an idea produces an artifact attached to that idea.
Some agents produce intermediate artifacts during longer tasks (the research notes that fed the final report, the outline that became the draft).

The intermediate ones are usually hidden behind a “Show research” or “Show working” link. The final ones are visible in the Artifacts tab like anything else.

The “draft a document by talking” pattern

A common workflow:

Tell your AI you want to produce something. Don’t over-specify; rough is fine.
It produces a draft artifact.
Read it. Push back on specific parts in plain language.
It revises. Repeat steps 3 and 4 until you’re happy.
Approve, export, send, or move on.

This is faster than writing from scratch and slower than “click a magic button”. The point is that the artifact is the thing you keep iterating on, not the chat history. The chat is the workshop; the artifact is the product.

When not to use an artifact

If the thing you want is a structured object — a contact, a list, a task, a calendar event — don’t make it an artifact. Ask for the object.

“Add Mike Reynolds as a contact. Roofing contractor. (555) 123-4567.” → contact, not artifact.

“Draft a profile of Mike Reynolds I can send to the project team.” → artifact.

The first creates a row in the contacts table. The second creates an editable document. Use the right tool.

Cleaning up

Old artifacts accumulate. Most can stay; storage is cheap and old context is sometimes useful. But if an artifact represents a draft you abandoned or a one-off you’ll never reference again, ask your AI to delete it. (Or batch them: “Delete every artifact older than three months that’s still in draft status.” It’ll show you what it found before doing anything irreversible.)