The app gives you two ways to ask the AI to write for you. Pick based on what you’re trying to do:

• Generate — when you know what shape you want. “A position paper on land use.” “Minutes for the spring assembly.” “A reply to the municipality’s request.” You describe it once, you get a draft back.
• Agent — when you’re still figuring out what you need. “What did we decide about X?” “Who keeps coming up in the housing meetings?” “What’s our current position on external partnerships?” You have a conversation until you’ve got what you need.

Both stay grounded in the cooperative’s real documents. Neither makes up something the archive doesn’t back up.

On the Generate page you fill in a small form: what kind of document, which cooperative it’s for (you can pick more than one if it’s a joint document), the prompt itself. Optionally you tick a few specific reference documents that should definitely feed the draft.

What the AI reads to write your draft:

1. The cooperative’s house rules — terminology, tone, things to never write, things to always include. Same rules every AI operation follows.
2. The leading documents of every cooperative in scope — statutes, regulations, brand story. The AI keeps the draft consistent with those.
3. The reference documents you picked — up to a handful, fed to the AI as additional context for this particular draft.

You see a preview with the AI’s suggested text before anything is saved. If it’s good — click Save as draft and the document goes into the pipeline as a draft awaiting approval. If it’s not what you wanted — click Discard and nothing is saved.

You can ask Generate to add a small policy footnote to the draft (one sentence pointing back at the cooperative’s public policy page) by ticking the box. Useful for documents that go outside the cooperative.

The Agent is a chat. You ask, it answers — but instead of pulling from general knowledge, it reaches into the cooperative’s own documents to find what’s relevant. You can keep asking follow-up questions in the same session; it remembers what you talked about.

When you start a session you choose how wide the agent should look:

• One organisation — the agent only sees that cooperative’s documents. Useful when you want a tight, focused conversation.
• All organisations you belong to — the agent can pull from any cooperative you’re a member of. Useful for cross-co-op questions.

While answering, the agent can:

• Search the documents — by keyword, type, language, or “only the leading ones”.
• Search by meaning — find documents that talk about a topic even if they don’t use your exact words. Needs the cooperative to have run the indexing step (admins do that).
• Open a specific document by name or id and read it in full.
• Look up the calendar — what meetings or deadlines are coming up, what already happened in a date range.
• Aggregate — “who is mentioned most across the housing minutes?”, “what laws come up most often?”, “which places appear across the harvest reports?”

Each answer the agent gives names which documents it consulted. You can click through to read the originals if you want to verify.

Every Agent session keeps the conversation history — your questions, the agent’s answers, and (for the audit trail) which documents the agent read along the way. You can come back to a session later and pick up where you left off. Delete a session and the messages are removed; the documents stay where they are. Platform admins can read any session for support and review.

• You know what kind of output you want, just need the words written → Generate.
• You don’t know yet what the answer looks like, you want to explore → Agent.
• You start with Agent to figure out the question, then move to Generate to draft the document → both, one after the other. Common pattern.