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

# Build a team brain

> Combine shared source-backed project knowledge with decisions and lessons that agents can recall in later sessions.

Use this when a team needs two kinds of continuity:

* project knowledge that people can trace back to sources;
* decisions, preferences, and lessons that agents should remember later.

This workflow uses Runtime for project knowledge and Memory for cross-session
context. Use Shapes only if the catalog cannot represent the team's records.

## Before you start

* Choose the shared project.
* Decide which source material belongs there.
* Agree on what deserves memory and whether its scope is agent, project, or commons.
* Exclude secrets and short-lived speculation from memory.

## Ask your agent

> Set up the Product Research project as our team brain for customer discovery.
> Check the project and use the best research Shape. Save the interview notes as
> sources, then stage the supported findings and open questions in one workspace.
> Show me the proposed records and their source passages before submission.
> After I approve them, remember the approved decisions and lessons at project
> scope. Verify the project records now and test memory recall in a new session.

## Steps

<Steps>
  <Step title="Check the project and research Shape">
    Inspect the active project. Search and add a catalog Shape only if the project does not already have a suitable one.
  </Step>

  <Step title="Save the team's source material">
    Add interviews, reports, and notes as sources. Retrieve a passage from each one before creating findings.
  </Step>

  <Step title="Stage the shared records">
    Open one named workspace for the batch. Add findings, open questions, and their links to evidence.
  </Step>

  <Step title="Review and submit the workspace">
    Read the workspace, fix validation errors, and compare representative findings with their source passages. Submit only after the team accepts the proposed records.
  </Step>

  <Step title="Save the operating memories">
    Store only approved decisions, stable preferences, and reusable lessons. Give each one an explicit kind and the narrowest useful scope.
  </Step>

  <Step title="Test both paths">
    Retrieve a project finding with its source evidence. Then start a new session and recall one of the saved decisions or lessons.
  </Step>
</Steps>

## Verify

From another connected agent or a fresh session, ask:

> What did we learn about enterprise procurement, what source supports it, and
> what team decision should I remember before the next interview?

The answer should identify which part came from project records, which passage
supports it, and which part came from memory.

## Troubleshooting

* **A finding has no source passage:** keep it as an open question or add the missing source before submission.
* **Workspace validation fails:** repair the current workspace and run validation again.
* **Recall returns irrelevant memories:** narrow the query, kind, scope, or time range.
* **A decision has changed:** archive the stale memory and save the replacement after review.

<Accordion title="Tool trace">
  ```text theme={null}
  Runtime: penumbra_status
  → penumbra_shapes_search / penumbra_shapes_add when needed
  → penumbra_add_source
  → penumbra_source_rag
  → penumbra_workspace_open
  → penumbra_capture
  → penumbra_workspace_read
  → penumbra_workspace_validate
  → approval
  → penumbra_workspace_submit
  → penumbra_omnisearch / penumbra_read / penumbra_source_rag

  Memory: memory_remember
  → new session
  → memory_recall
  ```
</Accordion>