Yes. Upload them to OpenAI Files (up to 1 GB per file) and the bot pulls cited passages at request time. A program officer asking about disbursement levels gets the real figure from the 990-PF rather than an invented one, and a journalist asking about three years of grants in a specific issue area gets a list with names, amounts, and years. No retraining is needed when you upload the next year's form or update the grants list.

Run two bots through multibot. The public bot reads eligibility, focus areas, and the public grants archive. The board bot is set to require logged-in users with a trustee user role, and its context includes governance documents and prior board minutes uploaded to a separate OpenAI Files vector store. Display conditions on user role and logged-in state keep the contexts cleanly separated, so no public visitor can ever query board-only documents.

It does not replace your LOI or application portal. The bot can describe what the application asks for, link the applicant to the right form, and answer questions about deadlines and the review cycle, but the actual submission happens in your existing system, whether that is a Gravity Form, a Fluent Forms intake, or a dedicated grants-management tool. Keeping the bot out of the submission flow avoids confusing accountability for a submitted application.

Yes. The system prompt can instruct the bot to describe general decline patterns at the program level (out-of-geography, out-of-focus, mismatch with current cycle priorities) without speculating about why a specific application was declined. Individual decisions are not in the bot's context, only published criteria and aggregate patterns, which keeps the bot useful without overstepping into territory only program staff should discuss.

Yes, when those are stored as structured content. Many foundations use ACF or custom fields for the next LOI deadline, the next full-proposal deadline, and the next board meeting date. The bot reads those fields and quotes the dates in conversation. Update the date in WordPress and every conversation reflects the new deadline immediately, without any prompt change or redeployment.

Inside your WordPress database. Nothing is shared with Sleek or any third party beyond the API calls your key authorizes. You decide retention, you decide which staff members can see logs, and you can purge old conversations on a schedule. The logs are particularly useful for spotting eligibility confusion patterns that suggest a rewrite of the focus-areas page.

The board bot's system prompt can include the conflict-of-interest policy verbatim and instruct the bot to refer specific COI questions to the corporate secretary rather than answering directly. The public bot has no access to internal COI matters at all. This separation matters for governance hygiene, and the multibot architecture enforces it at the data-source level, not just the prompt level.

Yes. After initial setup, the only ongoing maintenance is keeping the eligibility page, focus areas, deadlines, and grants list accurate, which the program team does anyway. The bot inherits those updates automatically because it reads from WordPress and OpenAI Files at request time. A program officer with editor access can adjust the system prompt and presets in under an hour a month, which fits comfortably within most small foundation operating cycles.