One agent. Your whole catalog.
The Feed is a grid where every row is a product and every column is data, an agent run, or a run’s output. Import your catalog, run a creative agent down every row, review the results in one drawer, and export a feed that Meta and TikTok pull on their own schedule.
Your catalog, cleaned on the way in.
Upload a CSV or connect Shopify. Orisu proposes a field mapping you confirm once, and later re-imports only touch rows that changed. New columns get a type inferred from your values, with an override right on the mapping screen.
Ask it to normalize your data and it writes small scripts per column, then shows you the real before and after over a sample of rows. Once you confirm, the script is saved on the column and replays on every future import.
Estimate first. Then every row.
Mount any published agent, or a community app, as a column. Orisu pre-matches the app’s inputs to your columns, and you test it on one real row before anything runs in bulk.
Run a cell, a column, or your selection. Batches above your cost threshold show an estimate and wait for you. Cells move from queued to running to needs review to approved, and a failed cell shows the error with a one click retry.
One drawer. Next, next, next.
Click any finished cell to see its outputs, approve, reject, or re-run, then jump straight to the next cell that needs a look. Approving is what makes a row eligible for export.
If the agent has a Human Review node, the run pauses at that checkpoint and the drawer shows what is waiting. Approve or reject each item and the run picks up from there.
A feed the channels pull themselves.
Pick Meta or TikTok, confirm the field mapping, and Orisu hosts a CSV at a stable URL. Paste it once into Commerce Manager or TikTok Catalog Manager and the channel refreshes on its own.
Every mapped field is checked against that channel’s real catalog spec before a row goes out. A row that fails a required check is blocked with the reason; an over-long title is truncated and flagged. Map with AI proposes the whole mapping and the compliance report for you.
Make the ad once. Run it two hundred times.
Import a catalog and watch the first column run. Free to start, and the free tier needs no card.
Feed questions, answered.
Does the Brain apply in a Feed run?
Yes. A Feed cell runs the same agent with the same Brain context as the studio, so every row gets your rules and references.
What happens if I edit the agent later?
A run column pins the agent's deployed version, so the feed keeps producing consistent results while you keep editing in the studio. Community apps show an update button and only change when you click it.
Can I use an app someone else published?
Yes. The agent picker lists Explore apps. Picking one saves a private copy to your workspace; runs bill to your credits and the copy never changes under you.
Can the copilot edit my feed?
Yes. It knows which feed is open and can add columns, set cells, estimate a batch, and write bulk transforms. Every write shows as a card you approve first.
Is there an API?
Everything the grid does is also available as MCP tools and on the REST API, so an external agent can build and run a feed end to end.
Run it down the whole list.
Import your products, mount one agent, and see the first column fill in today, free, no credit card.