How GrantLoop Works

A short tour of the pieces and how they fit together. If you read one page before starting, read this one — most early confusion comes from not knowing what the two spreadsheets are for.

One agent, three places you meet it

There is a single AI agent behind everything. You meet it in three places:

• Chat — the panel on the right. Ask for anything: find funders, write a draft, add a field, clean up a sheet. If the panel is hidden, a Chat Panel button brings it back.
• Sheets — spreadsheets of grant opportunities. The agent fills them in, keeps them current, and reads them when deciding what to apply for.
• Docs — the writing editor, where applications get drafted and edited.

The rule worth remembering: anything you can do by hand, you can ask the agent to do instead. It is usually faster and it teaches the agent how you like things done.

The two sheets every organization starts with

New organizations are set up with two sheets, and they do different jobs. This is the single most common point of confusion.

Potential Opportunities is the wide net. The agent adds anything that might be eligible, including things it later decides against. It is deliberately messy — think of it as the top of the funnel, and as a record of what has already been looked at so the same funder isn't researched twice. Rows marked low fit or sunset are not pursued; they are kept as history.

The Application Tracker is the record of real work. What was applied for, when, for how much, what came back, and when you can apply again.

So: prospects are candidates, the tracker is commitments. If a funder looks wrong in the prospect list, that's expected — it hasn't been judged yet.

These have been renamed a couple of times, so an older organization may have the same two sheets as Prospective Grant Opportunities and the Grant Applications Submission Tracker, or older still as Grant Opportunities and Grants Tracking. Same sheets, earlier names.

Why you might have more than two

Sheets accumulate. A duplicate gets created, an old year lingers, someone starts a separate list for a campaign. If you're looking at four sheets and can't tell them apart, ask the agent — "which of these sheets are duplicates, and which should I keep?" — and it will read them and tell you. You can then ask it to delete or merge the extras.

Having a separate sheet on purpose is fine and common: one for a capital campaign, one for a specific program, one per client if you're a consultant.

Two ways to look at a sheet

Every sheet has view tabs at the top:

• Table — the spreadsheet. Add fields, filter, sort, export.
• Board — the same rows as cards, grouped into columns by one field (usually status). Drag a card to change its status. Cards whose field is empty collect in No status.

Both views show the same data. Changing a view never changes the underlying rows.

A note on words, because they're easy to mix up: the columns of a table are fields; the columns of a board are groups, and each group is one option of the field the board is grouped by.

What the agent knows about you

Two separate things feed it, and they are not the same:

Reference Materials are files you upload — past applications, your 990, program descriptions, budgets. The agent reads them when drafting. This is the highest-leverage thing you can do: without them it only knows what is publicly findable about your organization, so drafts come back generic. Uploading a handful of old applications changes the quality immediately.

Memory is what it remembers about how you work — facts about your organization, and instructions like "we don't do after-school programming" or "always copy Catherine." You add memories by telling the agent in chat.

If you want something to stick beyond the current conversation, say so explicitly: "going forward, always do it this way." Otherwise it may treat the instruction as applying only to what you're working on right now.

Working while you're not there

Recurring Tasks are the agent working on a schedule rather than waiting for you. The common setup — often called autopilot — is a weekly run where the agent finds new opportunities, picks what to pursue, drafts an application, logs it in the tracker, and emails you the result.

Recurring Tasks is hidden in the sidebar by default. Turn it on in Settings with "Show Recurring Tasks in your sidebar."

What still needs you

The agent drafts; it does not submit on your behalf without you. Its edits to a document appear as an inline diff with Accept all and Reject all — nothing changes until you accept. Applications still need a human to confirm the ask amount, attach documents, and press submit. If a funder emails back, that's you too.