How it works
How Keeplore works
Last updated
Keeplore turns an interview into rules. Each rule has a condition, an action, a reason and the exact quote it came from.
Keeplore is knowledge capture for AI agents. In three steps: an AI interviewer chats with a person, a manager reviews the rules that come out, and your agent calls an API and gets the approved rules back.
What does Keeplore do?
- Interview. A manager adds an employee, who is leaving or staying, and sends a private link. Keeplore's AI interviewer chats with them in plain text.
- Review. Their answers become rules. A manager approves, edits or rejects each one.
- Query. Your AI agent calls
POST /api/v1/querywith an API key and gets an answer plus the approved rules and the name and role of the person each came from.
Only rules a manager has approved are ever returned to an agent.
How does the interview work?
A manager adds a person and sends a private link. Before the interview starts, the person sees a consent notice. It says managers at their company will see the answers, that each rule keeps their name as its source, and that they should not share passwords or personal customer details.
Then the AI interviewer chats with them in plain text, for about 30 to 45 minutes and up to 40 answers. The person can pause and come back. It is a text chat, not voice or video.
The interviewer asks one question at a time and works through six areas:
- A normal week
- Exceptions and edge cases
- Clients and partners
- Approvals and people
- Systems and workarounds
- What a new hire gets wrong
When someone says "usually", "normally", "it depends" or "except", it asks when the usual way does not apply and what happens then. If an answer is vague, it asks for a recent real case. It tells people not to share passwords or personal data about customers. Those are instructions to an AI, not a filter.
You choose per person whether they are leaving or staying, and the interview is worded to match. You can try it on yourself first.
What does a transcript excerpt look like?
Interviewer: Annual plans have a 30-day refund window. What happens after it?
Jonas: Officially it's no refunds on annual, but if they're past the window we do a prorated credit for the months left. Anything over five hundred I check with Dana in finance first.
Interviewer: Can you tell me about the last time that happened?
Jonas: Last month. A customer asked in week six and we credited the unused months.The employee said "officially", which signals an exception. The interviewer asked for a recent real case instead of moving on.
What is a rule?
A rule is a short card: a condition (when), an action (do), a reason (why) and the exact quote it came from. This one comes from the transcript above.
| Field | Value |
|---|---|
| When | A customer on an annual plan asks for a refund after the 30-day window |
| Do | Offer a prorated credit for the months left. If the credit is over €500, hand over to a person for finance approval |
| Why | Not stated. Keeplore leaves the reason empty if the employee did not give one |
| Quote | "Officially it's no refunds on annual, but if they're past the window we do a prorated credit for the months left. Anything over five hundred I check with Dana in finance first." |
| Source | Jonas Berg, Support Specialist |
Keeplore checks in code that each rule's quote appears in what the employee typed, and drops the rule if it does not. It looks only at the employee's messages, not the interviewer's. Very short quotes prove little. The check does not show the rule is right. That is what the manager's approval is for. How the quote check works.
How does the manager review rules?
A manager sees every rule with its quote and approves, edits or rejects it. Rules start as pending and nothing reaches an agent until a manager approves it.
If a new rule appears to contradict a rule you already approved, Keeplore flags it for the reviewer. It does not resolve the conflict. A person decides.
How does an AI agent get the answer?
Your agent sends a question to POST /api/v1/query with an API key. Keeplore looks up your approved rules, and a model writes a short answer from them. The response has an answer and the rules it used.
POST /api/v1/query
Authorization: Bearer lore_…
Content-Type: application/json
{ "question": "A customer on an annual plan wants a refund in week six. What do I do?" }{
"answer": "Offer a prorated credit for the months left. If the credit is over 500 euros, hand over to a person for finance approval.",
"rules": [
{
"id": "rul_8f2c41",
"topic": "Billing",
"condition": "A customer on an annual plan asks for a refund after the 30-day window",
"rule": "Offer a prorated credit for the months left. If the credit is over 500 euros, hand over to a person for finance approval",
"reason": null,
"source": "Example employee, Support Specialist",
"confidence": 0.8
}
]
}The response includes the person's name and role as source. It does not include the quote. Only the reviewer sees the quote.
confidence is a number from 0 to 1. It reflects how certain the employee sounded in the interview. It is not a measure of accuracy.
If no approved rule applies, the answer is empty. Your agent can treat an empty answer as "hand over to a person". When a rule does apply, a model writes a short answer from the approved rules, and the rules it used come back with it so your agent, or you, can check. The answer text itself is model-written and is not checked against the rules.
Limits
Every plan has a monthly AI budget. When it is reached, POST /api/v1/query returns HTTP 402 with "This workspace has reached its monthly AI limit." Each API key is limited to 60 queries a minute. Trial workspaces can call the query API.
Export
GET /api/v1/export returns every approved rule as JSON. Add ?format=md for Markdown. You can load either into your agent's guidance or knowledge feature. There is no pre-built integration with agent platforms today.
What does Keeplore not do?
- It has no integrations with email, Slack or calendars, on purpose. Keeplore uses only what people say in the interview.
- It has no SOC 2 or ISO certification today.
- It is not a chatbot for your staff. The interview is a one-time conversation to capture knowledge. Your agent is what answers questions.
- It does not prove a rule is right. A manager decides.
How is my data handled?
The app and database run in Frankfurt. Interview text is processed by Anthropic in the United States, sign-in data by Clerk in the United States, and invitation emails go through Resend in the United States. Employees give consent before an interview starts, and you can delete an interview and its rules in one click. Details are on /security.
Related
01What happens when Keeplore doesn't know the answer?
If no approved rule applies, the answer is empty and so is the list of rules. Your agent can treat that as "hand over to a person".
02Can Keeplore interview people who are staying?
Yes. You choose per person whether they are leaving or staying. You can also try it on yourself first.
03How do I get the rules into my own tools?
Call GET /api/v1/export for every approved rule as JSON or Markdown and load it into your agent's guidance or knowledge feature, or have your agent call POST /api/v1/query.
Try it on one person.
Pick someone your agent is missing knowledge from. Send them the link. Read the rules that come back.