Knowledge & Memory

Ask questions of your own documents

The fastest win in the product, and the foundation most other playbooks build on.

Time
6 minutes
You work in
Documents
Connect first
Works out of the box

Before you start

  • A handful of documents worth asking about

This is the fastest useful thing you can do in Tenfold, and most of the other playbooks assume you have done it.

The idea is narrow on purpose. A general-purpose model knows a great deal about the world and nothing at all about your refund policy, your MSA, or the reason you switched suppliers in March. Documents is where you fix that, once, for every employee you will ever hire.

Upload something worth asking about

Open Documents and drag in a handful of files. PDFs, Word documents, plain text and markdown all work.

Resist the urge to upload everything. Five documents you actually reference beat five hundred you do not, and they make it much easier to tell whether the answers are good. Contracts, policies, a handbook, past proposals and onboarding notes are all high-value starting points.

The shared library, and what an employee pulled out of a contract

Each file is parsed and indexed as it lands. Larger PDFs take a moment.

Ask a question you already know the answer to

This is the step people skip, and it is the one that tells you whether to trust the thing.

Use Ask, and ask something you can verify yourself in ten seconds. Not "summarise our policies" but "how many days notice does the Acme contract need for termination". You are not testing whether it can write. You are testing whether it read your document or produced a confident average of every contract on the internet.

Prompt
According to our documents, how much notice do we have to give to
terminate the Acme agreement, and which clause says so?

Always ask for the clause, the section or the document name. An answer that cites its source can be checked in seconds. An answer that does not has to be trusted, and you have no reason to trust it yet. Asking "and where does it say that" turns a plausible answer into a verifiable one.

Ask something you do not know

Now the actual value. Ask across documents, which is the thing you cannot do by opening files one at a time.

Prompt
Look across all of our supplier contracts and tell me which ones
auto-renew, when each renewal window opens, and which ones we would
have already missed if we did nothing this month.

This is a genuinely different operation from search. There is no single document that contains that answer, and no query that would surface it.

Point an employee at it

Documents are workspace knowledge, not a private feature of the Ask box. Any employee you hire can search them, which is what makes the rest of the library work: the support playbook answers from your help docs, the recruiting playbook scores against your requirements document, and neither needs you to paste anything into a prompt.

If you want an employee to specialise, put its material in its own folder and say so when you brief it.

What good looks like

You should be able to ask a question that would have taken you fifteen minutes of opening files, and get an answer with a source you can check in ten seconds.

If answers come back vague, the usual cause is not the model. It is that the document does not actually say what you thought it said, which is worth knowing on its own.