> ## Documentation Index
> Fetch the complete documentation index at: https://docs.northernlogic.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Get your first useful AI answer

> Save a sample policy, verify a source-backed answer, and prove an update reaches your tool.

**Outcome:** your AI tool answers from a saved page and uses a changed fact
after you update the source.

**Before you start:** use a test page in a vault where you have Editor or Vault
admin access. Connect one tool through
[local sync or MCP](/getting-started/connect-ai). The policy below is fictional
training content.

## 1. Save a small source

In **Knowledge**, choose the test vault and create `practice-client-policy.md`
with **+ File → New**. Open **Edit**, paste the following, and select **Save**.

```md theme={null}
# Practice client policy

Fictional training example. Not an approved company policy.

## Meeting preparation

The account owner sends the agenda two business days before the meeting. The
agenda includes open decisions and links to the project brief.

## After the meeting

The account owner records decisions and next actions in the project brief.
```

Reopen the page and search for `Practice client policy`. If using a personal
vault, select it explicitly in Search.

## 2. Ask your connected tool

Copy this prompt and replace the vault description with the one you chose:

```text theme={null}
Use Northern Logic Knowledge. Orient to my workspace and find
"Practice client policy" in my test vault. Open the saved page.
When should the account owner send the agenda? Include the source
path and the sentence that supports your answer. Do not edit files.
If you cannot access the source, say so instead of guessing.
```

**Expected result:** two business days before the meeting, with the path to your
test page and a supporting sentence. Open that source yourself to confirm it is
the intended page.

## 3. Update the source

In the web editor, change **two business days** to **three business days** and
save. For a local tool, wait for the replica to receive the change and
[check sync health](/connections/sync-health).

```mermaid theme={null}
flowchart TD
  P[Save the policy] --> Q[Ask a question]
  Q --> V[Open and verify the cited source]
  V --> E[Edit the saved policy]
  E --> F[Fetch the page again]
  F --> C[Confirm the changed answer]
```

## 4. Ask for a fresh read

```text theme={null}
Fetch "Practice client policy" again from Northern Logic Knowledge.
Do not rely on your previous answer. What is the current agenda
lead time? Include the source path and supporting sentence.
```

**Expected result:** three business days. A new chat can help separate this
check from earlier conversation context, but the critical step is actually
fetching the current source.

## Check your result

* The tool used the correct workspace and vault.
* Both answers identify the same test page.
* The second answer reflects the saved change.
* **Activity** shows your edit; the read-only exercise did not ask the agent to
  change anything.

## If it did not work

| Symptom                       | Next check                                                                                         |
| ----------------------------- | -------------------------------------------------------------------------------------------------- |
| No source or an invented path | Ask for the actual tool read; follow [agent search troubleshooting](/help/agent-cannot-find-page). |
| It still says two days        | Confirm Save completed, then inspect the replica or request a fresh MCP read.                      |
| It finds a different policy   | Specify the exact vault and path returned by orientation.                                          |
| It can read but cannot write  | Reading worked. Writing has separate [permission requirements](/collaboration/vault-access).       |

## Use this in real work

Replace the practice source with an approved procedure and repeat the evidence
check. Keep the practice page clearly labeled or remove it when finished. Model
output still needs human review.

**Next:** [Write in your company’s voice](/tutorials/company-writing) or
[Use the same knowledge across AI tools](/tutorials/shared-ai-context).