# Teach Connect about your business

Knowledge is the material your answers are grounded in and it applies to everybody. Memory is what Connect has learned about your business and its people, held at four tiers and applying only as widely as the tier it sits in. Almost every early mistake is a fact about one customer written into Knowledge, where it silently starts applying to all of them.

- **Status:** Available
- **Audience:** both
- **In the app:** #/knowledge, #/knowledge-facts-advanced
- **Last verified:** 2026-09-10
- **Canonical:** https://connectbyjbrh.com/docs/getting-started/teach-connect-your-business/

## The distinction, in one table

|  | Knowledge | Memory |
|---|---|---|
| What it is | Source material a workspace supplies | What Connect has learned and been told |
| Who it applies to | Everyone | Only the tier it is stored at |
| Tiers | One — the workspace | Workspace, channel, endpoint, contact |
| Typical entry | 'These are our delivery terms' | 'This customer prefers WhatsApp' |
| Who can remove it | Anyone editing the source | Anyone, at any tier, from the memory viewer |
| Where it shows up | Grounded answers and extracted facts | The instructions for one piece of work |

## The first four things to write down

1. **What you sell**, in the words customers use for it rather than your internal product codes.
2. **What you do not do.** The highest-value entry there is, because it is what turns a confident wrong answer into an honest 'we don't do that'.
3. **Your terms**: delivery, lead times, warranty, returns — whatever people ask about most.
4. **How you answer**: the tone, who signs, and anything you always or never say.

Add the first three as Knowledge sources. The fourth is a behaviour rather than a fact and belongs on the Connect Rules screen as a standing instruction, which is a durable direction rather than something to be quoted back to a customer.

## Grounding has a budget

Adding more material does not put more material into every answer. Grounding is selected and budgeted before it reaches the model, and on a live phone call the budget is tight and measured: roughly 2,000 characters of knowledge and no more than four facts, about 700 characters of memory, and around 600 for the contact's own block. There is a direct reason for the ceiling — a longer prompt measurably delays the first word spoken on a call, and on a phone line that delay is the difference between a conversation and a pause.

The practical consequence is that quality beats volume. Four precise sentences about your returns policy do more than a forty-page handbook, because the four sentences will be selected and the handbook will be sampled. [What every character of a prompt costs on a live call](/research/prompt-size-and-first-token/) has the measurements.

## What belongs in Memory instead

**A person's preference** — Contact tier. 'Always rings rather than emails', 'do not call before ten'.
**A quirk of one mailbox or line** — Endpoint tier. 'This address is monitored by the warehouse, keep replies short'.
**Something true of a whole channel** — Channel tier. 'On WhatsApp, keep it to two sentences'.
**A standing fact about the business** — Workspace tier — or Knowledge, if it is material an answer should quote.
**Never contact this person again** — A block directive on the contact. It is stored as a tag rather than as prose, which is what makes it hold across every channel and every future conversation.

Everything Connect remembers can be read at every tier and forgotten from the same place — the memory viewer opens from the mailbox row, the channel screen and the contact panel. An empty tier is shown as empty rather than hidden, so 'nothing is set here' is a visible answer instead of an assumption.

## Questions

### Can I just paste my website into Knowledge?

You can, and it is a reasonable start, but it is rarely the best material. Marketing pages describe what you want to be true; customers ask about lead times, exceptions and edge cases. The entries that change draft quality most are the ones that answer questions you are tired of answering.

### If I correct a draft, does Connect remember the correction?

Not by itself. Editing a draft changes that message; the edit is what gets sent, and Connect does not write over it. For the correction to change future behaviour it has to be written down — as a standing instruction if it is about how to write, as Knowledge if it is a fact, as Memory if it applies to one person or one line.

### How do I check what Connect currently knows about someone?

Open the memory viewer from that contact and read the tiers. It shows what applies to this person, what applies because of the channel or endpoint, and what applies workspace-wide — separated, so you can tell which level to change.

## Related

- [Knowledge in Connect](https://connectbyjbrh.com/docs/knowledge/)
- [Memory in Connect](https://connectbyjbrh.com/docs/memory/)
- [Your first week](https://connectbyjbrh.com/docs/getting-started/first-week/)
- [Check what Connect knows about a person](https://connectbyjbrh.com/docs/how-to/check-what-connect-knows/)
- [Give Connect a document to answer from](https://connectbyjbrh.com/docs/how-to/give-connect-a-document/)
- [What every character of a prompt costs on a live call](https://connectbyjbrh.com/research/prompt-size-and-first-token/)

## What this page is based on

- docs-source/sources/GENERAL.md §6 — memory tiers and directives
- Connect capability registry (docs-source/facts.py) — MEASURED grounding budgets
- docs-source/sources/GENERAL.md §4 — knowledge modules and screens