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

# Knowledge sources

> Build Knowledge from a website, route uploaded files, inspect grounding, and verify an important fact in a fresh task.

**Knowledge** is the workspace surface for approved source material and the artifacts grounded in it. Use it for brand facts, reference evidence, and source-backed research—not for reusable operating procedures, which belong in **Skills**.

Open [Knowledge](https://app.marketer.com/dashboard/knowledge-base) and confirm the active workspace.

## Understand the landing page

| Area                 | What it answers                                                 | What to inspect                                                         |
| -------------------- | --------------------------------------------------------------- | ----------------------------------------------------------------------- |
| **Standing index**   | What compact context is available to agents on future requests? | Current facts, contradictions, source authority, and missing essentials |
| **Recent artifacts** | What reports or analyses did the workspace produce recently?    | Type, title, summary, date, detail, and **Grounded in N** sources       |
| **Sources**          | Which inputs feed workspace context?                            | Route, owner, date, processing state, and authority                     |

The landing page is intentionally a slice: **Recent artifacts** prioritizes a limited set, and the **Standing index** is compact context rather than the entire knowledge base. Open the artifact or source detail for a decision-grade review.

## Choose the right route

| Material                     | Select                   | Resulting purpose                           |
| ---------------------------- | ------------------------ | ------------------------------------------- |
| Approved brand identity      | **Brand kit material**   | Brand Profile and enduring brand context    |
| Factual or reference source  | **Reference knowledge**  | Evidence agents can retrieve and cite       |
| Repeatable method            | **Working procedure**    | A Skill to govern and test                  |
| Truly mixed or unclear input | **Let the agent decide** | A proposed route you must inspect afterward |

Resolve mixed files before upload when possible. A document that combines confidential notes, outdated claims, brand identity, and a procedure is difficult to govern as one source.

## Build from a website

Use website research when the approved public brand or storefront site should establish current workspace context.

<Steps>
  <Step title="Enter the authoritative site">
    In **Sources**, provide the canonical domain. Avoid a campaign mirror, unrelated retailer, or regional site that does not own the facts you need.
  </Step>

  <Step title="Select Build from website">
    Confirm the exact domain before starting. The source moves into **Brand research running** while Marketer gathers and structures available material.
  </Step>

  <Step title="Open view progress">
    Follow the active task and its visible steps. Closing the task does not mark the source complete.
  </Step>

  <Step title="Inspect the new context">
    Return after a terminal state and review the source, standing context, and recent artifacts. Check product claims, audience conclusions, dates, and brand details against the current site.
  </Step>

  <Step title="Test one sourced fact">
    Start a fresh task, ask a question with one verifiable answer from the site, require the source and effective date, then compare the response with the live page.
  </Step>
</Steps>

A website build reflects what was available during that run. Rebuild or replace it after a material site change; do not assume every later request performs a new live crawl.

## Upload and route files

The upload surface accepts approved PDFs, Markdown, text, images, and fonts, up to the **100 MB** limit displayed by the UI.

<Steps>
  <Step title="Set Route uploads as">
    Choose **Let the agent decide**, **Brand kit material**, **Reference knowledge**, or **Working procedure** before adding files.
  </Step>

  <Step title="Add the exact files">
    Use **Drop files or click to upload**. Confirm filenames, remove duplicates, and ensure every file opens and contains selectable, understandable content when text extraction is expected.
  </Step>

  <Step title="Read the source state">
    A row can show upload or processing progress before **Done**, or finish as **Failed**. Do not test retrieval while it is still processing.
  </Step>

  <Step title="Verify the resulting route">
    Under **In your knowledge base**, read the arrow destination such as **Brand kit**, **Knowledge**, or **Skill**. The resulting badge, not the original intent, confirms where the source landed.
  </Step>

  <Step title="Retry only the failed row">
    Select **Retry** after checking file integrity and type. Avoid uploading another identical copy while the original can be recovered.
  </Step>
</Steps>

Identical file bytes can be recognized and reused rather than reprocessed as a distinct source. Check existing rows before assuming a repeated upload created new context.

Images and fonts are treated as brand assets rather than ordinary text documents. Verify their Brand Profile effect and rights separately from a factual retrieval test.

## Change a source destination

Use the row's route control when a source landed in the wrong place.

* Moving a source into **Brand kit** can start metered re-reading; review the confirmation and displayed usage.
* Moving a source out of **Brand kit** changes future routing, but details already extracted into the Brand Profile can remain.
* Moving a procedure to **Skill** creates behavior that needs trigger, scope, body, and negative-trigger review.
* Reopen both the source and destination after processing completes.

Do not interpret rerouting as deletion. Inspect already-produced brand context and artifacts before deciding whether further cleanup is required.

## Review artifact grounding

Open a row in **Recent artifacts** and inspect its full detail.

Confirm:

* Artifact type, title, summary, and date
* The evidence named by **Grounded in N**
* Which claims come directly from a source
* Which statements are synthesis or recommendation
* Whether a newer or higher-authority source conflicts

An artifact summary is orientation. The open artifact and its sources are the review surface.

## Validate Knowledge in a fresh task

<Frame caption="Task progress can show that Marketer searched and read saved workspace context; verify the final fact against the named source.">
  <img src="https://mintcdn.com/marketer-com/Zy37cwRIaZeFRNpC/assets/product-flows/task-progress.png?fit=max&auto=format&n=Zy37cwRIaZeFRNpC&q=85&s=63fd4b25b6d98ea3eed544364fb2e9a0" alt="Marketer task progress showing saved brand memory in search context and a completed step reading saved brand memory" width="1468" height="879" data-path="assets/product-flows/task-progress.png" />
</Frame>

1. Start a new task so prior conversation context does not mask the test.
2. Ask one specific question the source can answer.
3. Select **+** → **Add context**, then attach the relevant product, brief, campaign, or creative when it is part of the question. Use `@` only when you want to mention an agent.
4. Require the source title, cited fact, effective date, and any conflicting source.
5. Compare the answer with the uploaded file or live website.
6. Correct the source, route, or authority order and retest if the result is stale or ambiguous.

## Completion check

* Every source has an owner, date, authority, and intentional route.
* Website and upload work reached a terminal state.
* Failed rows were fixed or deliberately removed.
* Brand kit, Knowledge, and Skill routes match the material's purpose.
* Important artifacts identify their grounding.
* A fresh task retrieved a known fact and named the correct source and date.
* Conflicting or outdated authorities were resolved.

## Troubleshooting

<AccordionGroup>
  <Accordion title="Brand research stays in progress">
    Open **view progress**, inspect the failed or waiting step, and confirm the canonical domain is reachable. Resolve the existing task before starting another identical build.
  </Accordion>

  <Accordion title="A file fails">
    Confirm it opens locally, fits the displayed size limit, and uses a supported type. Use **Retry** on the exact row after correcting a malformed or unreadable export.
  </Accordion>

  <Accordion title="The file went to the wrong destination">
    Use the row's route control and review the move confirmation. After processing, inspect both the new destination and any previously extracted Brand Profile detail.
  </Accordion>

  <Accordion title="The agent ignores a source">
    Confirm the row is **Done**, the route is correct, and a higher-authority source does not contradict it. Test in a fresh task with the source named explicitly.
  </Accordion>

  <Accordion title="An artifact is grounded in stale evidence">
    Replace or rebuild the outdated source, resolve conflicts, and rerun the decision-grade task. Existing artifacts preserve their earlier context and should not be silently relabeled as current.
  </Accordion>
</AccordionGroup>

## Continue

<CardGroup cols={2}>
  <Card title="Govern Skills" icon="list-check" href="/product-guides/knowledge-skills">
    Turn approved procedures into scoped, testable agent behavior.
  </Card>

  <Card title="Use context in chat" icon="at" href="/working-with-marketer/chat-and-context">
    Attach the exact record and verify how the Marketing Agent uses it.
  </Card>
</CardGroup>


## Related topics

- [Knowledge and Skills](/product-guides/knowledge-and-skills.md)
- [Skills](/product-guides/knowledge-skills.md)
- [Run your first task](/get-started/first-task.md)
- [Navigation and Quick actions](/working-with-marketer/navigation-and-quick-actions.md)
- [Context and grounding](/essential-guidelines/context-and-grounding.md)