field

FAQs

Plain answers to plain questions.

What is an AI helper, exactly?v

An AI model (Kimi K2.6 in v0) that lives in your room. It reads the cards you write. It can add witness cards from a different model family than the one that wrote your text. It cannot edit, rewrite, or delete what you wrote. It is the "second pair of eyes" that sticks around.

What does "remembers" mean? Is it really memory or is it just sending old text back into the prompt?v

Both. The room is a database. Every card and every edge is a row that persists. When the helper responds, it reads from that database. Unlike a chat session, the row does not disappear when you close the tab. You can come back in three weeks and it is still there.

Can the AI helper change something I wrote?v

No. By kernel rule, your cards are append-only. If something needs revision, you write a new "revision" card. The original row stays. The system surfaces both. You can compare them.

What are the "five tools" the landing mentions?v

Talk (the helper writes back), Browse (search inside the room or web), Buy (a market for AI capability upgrades), Build (generate drafts and artifacts), Map (a placeholder; the network of rooms, coming soon).

Is this a Twitter clone?v

No. Twitter is a feed of isolated posts optimized for circulation. Field is a feed of rooms. Each room is a project. Inside the room, work compounds. You return to a room because there is something to come back to. You return to Twitter because something new might appear.

Is this a learning app?v

You can use it that way. The structure (cards, edges, revisions, return prompts) is well-suited to deliberate practice and spaced revision. But there is no curriculum imposed from above. You and your network produce what gets learned.

What is "polycosm"?v

The kernel theory underneath field. A polycosm is a network of rooms, each room a small conserved world. You can read the paper at /paper. You do not need to read it to use field.

How is this different from Notion or Google Docs?v

Notion and Docs are open canvases. You decide what gets stored. Field has typed cards (claim, question, source, revision, etc.) and typed edges (supports, refutes, refines, answers). The structure makes the room readable to a witness, not just to you.

Is the AI in the room going to commit to things on my behalf?v

Never without your explicit, named action. There are no auto-purchases, auto-promotions, or auto-decisions in field. Every irreversible row is named-author and gated by an explicit button press.

Can I delete a room?v

You can archive a room. By kernel rule, the room and its cards persist after archive. If you need a hard delete (e.g., you typed something private into a public room), use the "redact" action which appends a redaction record over the row. The original row is replaced with a redacted version; the redaction event itself is logged. We do not silently destroy authorship.

What does it cost?v

The early-access cohort is free. Public pricing is gated on the Mirofish launch simulation passing its preregistered F-claim contract. Until then, every seat is invited.

Why two AI substrates? Why not one big model?v

Per the Polybrain finding (RCC-n30, 2026), a single substrate cannot witness itself. We need a model from a structurally different family to verify the work of the first. Field uses Kimi K2.6 (Moonshot family) for the in-room helper and Gemini (Google family) for the math witness on the kernel paper.

Who built this?v

crest. Two co-founders: Andy Salvo and Jameson Ackerman. crest is the umbrella; field is the first product. The kernel theory underneath is the Polycosm paper (preprint at /paper).