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

# Org Brain

> Give support agents and your team access to permitted organizational knowledge, cited evidence, and reviewed lessons from completed work.

Org Brain is the knowledge layer for your [support agents](./digital-workers) and the people working alongside them. Use shared operating facts and earlier findings to guide investigation, understand an approved recovery method, or prepare an informed customer response. It indexes eligible Studio content, retrieves evidence within the caller's permissions, and can generate cited answers for human members.

Saved [memories](./memories-and-search) are one source. Org Brain also supports eligible conversations, text files, worker reports, and facts extracted by the dream cycle. It does not make every Studio record public or turn a remembered fact into live system evidence.

In the [full support loop](./support-loop), an agent searches permitted organization-visible evidence, checks the live system, and works toward the support outcome. People review the result and decide which lessons should become shared knowledge for later cases. Workers do not retrieve channel-only or private sources.

```mermaid theme={null}
flowchart TD
    accTitle: Org Brain knowledge flow
    accDescr: Eligible sources enter the index with their visibility preserved. Human members can retrieve authorized organization and channel evidence, with optional cited answer synthesis. Workers can retrieve organization-visible raw evidence only and check it against live systems during investigation.
    Sources[Eligible Studio sources] --> Index[Index with source visibility]
    Index --> Human[Human member<br/>Authorized org and channel evidence]
    Index --> Worker[Worker<br/>Organization-visible evidence only]
    Human --> Answer[Raw search or cited answer<br/>Synthesis requires consent]
    Worker --> Investigation[Raw search for investigation<br/>Check against live systems]
```

## Enable Org Brain

Org Brain requires an eligible organization on manual or Enterprise pooled billing, at least two organization members, and the organization owner's consent. Administrator access alone does not grant owner consent.

<Steps>
  <Step title="Open Organization settings">
    In the web administration console, open **Organization** for the correct organization. Find **Org Brain** and choose **Request activation** as the organization owner.
  </Step>

  <Step title="Wait for Active">
    Read the consent dialog and submit the request. **Activation requested** means setup is still pending. Use **Refresh** to check progress; contact Altostrat if the request remains pending. Search becomes available when the status is **Active**.
  </Step>

  <Step title="Enable generated answers">
    In **Org Brain answer synthesis**, review the displayed providers, models, and processing locations. Choose **Enable answer synthesis** to consent to the current policy. Search and answer synthesis have separate controls.
  </Step>

  <Step title="Check your conversation's tools">
    Use a current Studio desktop release. The active channel and your organization's tool policy must allow **Org Brain**. Start with a Personal conversation, or a shared channel whose Org Brain capability is enabled.
  </Step>
</Steps>

Saved synthesis consent can become inactive after a lifecycle or provider-policy change. Refresh the settings and review the current policy before enabling it again. See [organization administration](./organization-administration) for owner and administrator responsibilities.

## Ask a question and check its sources

Ask a specific question, such as:

> Ask Org Brain what our team decided about the Melbourne router maintenance window, and cite the sources.

Copilot uses `ask_org` for a generated answer. The **Org Brain answer** card contains inline evidence markers such as `[E1]` and a **Sources** list. Compare consequential claims with the evidence before acting.

Memory citations are interactive: select a marker or its source title to open the current authorized Memory record. Other source types are identified in the Sources list but do not all have a direct-open action in the answer card.

<Note>
  A Memory link opens today's saved source. It may have changed since the answer was generated. After editing a source, ask again. After deletion or loss of access, an old link can show **This source is no longer available** while the earlier answer remains in chat history.
</Note>

`search_org_knowledge` is the separate raw-evidence search tool. It returns matching snippets and citations without a synthesized answer. An unavailable or denied answer is not a reason to bypass its policy through another search path.

### Try it with one shared Memory

1. Ask Copilot to save a short, clearly titled fact **for the organization**, with no secrets or sensitive test data.
2. Review and approve the organization-wide save. Confirm the tool reports that it saved the record.
3. Open a fresh conversation and ask Org Brain a bounded question about that fact.
4. Inspect the cited Memory and check the answer against it.

Saving and indexing are separate steps. A new or edited source may take time to become searchable; there is no guaranteed instant indexing interval. Repeated requests can consume additional AI usage.

## What can become evidence

| Source | Available scope and limits |
| - | - |
| Organization-visible memories | Shared facts and work-product records. Updates and deletions feed the index. |
| Eligible shared conversations | Messages and summaries retain their organization or channel scope. A private conversation is not made team-visible by indexing. |
| Finalized text files | Supported text, Markdown, CSV, JSON, and code. PDF, image, binary, and oversized content can be skipped. Organization-visible standalone files can supply shared evidence. |
| Worker run reports | Completed and failed run reports stay within the worker's bound channel. Authorized channel members can retrieve them; worker search cannot retrieve channel-only reports. |
| Active knowledge facts | Facts published by the dream cycle or approved by an administrator, with source provenance. |
| Call transcripts | Transcript-source activation and owner-only search are not generally available. Do not assume a saved transcript is searchable in Org Brain. |
| Terminal recordings | Automatic cloud archival remains disabled by default; recordings are not an automatic shared knowledge source. |

File indexing is narrower than ordinary file access. Private, inherited, and chat-attached files are treated as owner-only; files shared with selected people and Memory attachments are excluded from this feeder. Owner-only indexing does not mean private Org Brain search is enabled. Private-source retrieval remains separately restricted.

Source-by-source admin switches are not yet a general self-service control. Contact Altostrat about source availability rather than assuming every source listed in a registry or roadmap is enabled.

## How permissions affect answers

Org Brain checks the active organization, caller, source visibility, and channel membership on the server. Human members can retrieve authorized organization and channel evidence. A worker can search organization-visible evidence only. Its bound channel must permit the search tool, but that permission does not expand results to channel-only or private content. Workers use raw search, not answer synthesis.

An organization administrator does not automatically gain Org Brain access to a channel they have not joined. A shared conversation in a team channel stays channel-scoped. A Personal conversation explicitly shared with the organization can become organization-wide evidence.

Shared chat replies also exclude personal Memory context that other readers should not receive. Use a Personal conversation for your own private context; moving to a shared audience does not grant teammates access to it.

## Review the organization's knowledge

Open **Administration → Org Brain** in the web console.

| Tab | What to do |
| - | - |
| **Review queue** | Inspect lower-confidence proposed facts and approve or reject them. Pending facts are not used in answers. |
| **Consolidation** | Compare duplicate or conflicting shared memories. Choose whether to retire the redundant record or keep both. Retiring requires confirmation and deletes that Memory. |
| **Facts** | Browse facts by state, inspect provenance, and retract an active fact that should no longer inform answers. |
| **Dream runs** | Inspect run status, processing counts, failures, and reasons a job was skipped. |
| **Dream settings** | Choose the organization's timezone and run hour, enable or disable each job, and edit its guidance. |
| **Registry** | Inspect organization-visible indexed documents and their source/status facets, including skipped or redacted results. |

The Registry omits private and channel-scoped documents and does not run semantic search. An empty Registry does not prove that a permitted channel has no indexed evidence.

### Dream cycles

The **Knowledge writer** extracts durable facts from eligible organization-visible documents. Facts with at least **85% model confidence** can publish automatically; lower-confidence facts wait for review. Confidence is the model's assessment, not a guarantee of correctness. Administrators can retract published facts.

The **Consolidator** proposes retiring redundant or contradictory shared memories. It waits for an administrator's decision before changing a record.

The two jobs have different review paths. Publishing a knowledge fact does not change an agent's workflow or authorize an action.

```mermaid theme={null}
flowchart TD
    accTitle: How dream-cycle results become shared knowledge
    accDescr: Organization-visible sources feed the Knowledge writer and Consolidator. The writer can publish facts at or above 85 percent model confidence; lower-confidence proposals need administrator approval. Consolidation always needs an administrator decision before a Memory is retired.
    Sources[Eligible org-visible sources] --> Writer[Knowledge writer]
    Sources --> Consolidator[Consolidator]
    Writer -->|At least 85% model confidence| Fact[Published fact]
    Writer -->|Lower confidence| Review[Admin reviews proposal]
    Review -->|Approve| Fact
    Review -->|Reject| Unpublished[Not published]
    Consolidator --> Decision[Admin compares memories]
    Decision -->|Confirm retirement| Retire[Retire selected Memory]
    Decision -->|Keep both| Keep[Memories unchanged]
```

In **Dream settings**, set **Timezone** and **Run at**, use the per-job switches, or choose **Edit guidance**. Guidance changes what the job looks for; it does not remove the service's security rules. Check **Dream runs** when no results appear: lack of activity, disabled jobs, provider policy, or machine-spend limits can skip a run.

[Worker improvement proposals](./digital-workers#outcomes-and-improvement) are reviewed separately in the agent's Outcomes view. They do not silently rewrite its workflow.

## Updating, disabling, and deleting

* **Edit a source** to correct the underlying fact, then ask again once the update is indexed.
* **Delete a source** to remove its indexed evidence through the deletion pipeline. Propagation is asynchronous; retained historical answers are separate chat content.
* **Retract a learned fact** in **Facts** when that derived claim should stop being used. Removing one source does not necessarily remove a fact supported by other sources.
* **Disable answer synthesis** to stop new generated-answer attempts while leaving Org Brain search under its existing setting. An attempt already running can finish.
* **Disable Org Brain** to turn off its indexing and search. This is not a complete data-erasure request; saved source records and historical chat remain.

Organization offboarding and member erasure use separate administrative processes. See [security and privacy](./security-and-privacy).

## Troubleshooting

| What you see | Next step |
| - | - |
| **Activation requested** | Refresh Organization settings and contact Altostrat if setup remains pending. |
| Org Brain tool is missing | Update Studio and check organization, channel, and member tool policy. |
| Answers are not enabled | The owner should inspect the separate synthesis consent and current provider policy. |
| Not enough evidence | Narrow the question, check source scope and indexing, and verify that the source exists. This does not prove the underlying fact is absent. |
| Couldn't verify this source | Retry the source inspection; a connection failure is different from an unavailable source. |
| Source is no longer available | Check whether the source was deleted or access changed. Re-ask using current permitted evidence. |
| New content is missing | Inspect Registry status and any indexing-budget notice; the Registry only shows org-visible documents. |

<CardGroup cols={2}>
  <Card title="Memories and search" icon="brain" href="./memories-and-search" arrow="true" cta="Save useful facts">
    Choose private or organization-wide Memory and keep work-product records current.
  </Card>

  <Card title="Digital workers" icon="bot" href="./digital-workers" arrow="true" cta="Use knowledge in a workflow">
    Run channel-scoped agents, review approvals, and inspect their outcomes.
  </Card>
</CardGroup>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.