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

# Data lifecycle

> One-page map of what Kair stores, what it never stores, how long it keeps it, and what erasure and export do at personal versus organisation level.

This page is the overview. It links to the pages that hold the mechanics: [Privacy & data ownership](https://docs.kair.is/platform/privacy.md), [Data subject requests](https://docs.kair.is/platform/data-subject-requests), [Exports](https://docs.kair.is/platform/exports.md), [Account ownership](https://docs.kair.is/platform/account-ownership).

> [!NOTE]
>
> Kair does not measure how much data a workspace holds, so this page describes **what** is kept and **for how long**, not volumes. Audio is by far the largest item.

***

## Where personal data goes and where it is cleaned

```mermaid theme={null}
flowchart LR
    A[Audio from device, browser,<br/>phone or meeting bot] --> S3[(S3, SSE-C encrypted<br/>original audio kept)]
    S3 --> T[Transcription worker]
    T -->|"NER + regex redaction<br/>people become PERSON,<br/>emails, phones, IBANs removed"| TX[(Postgres<br/>transcript segments)]
    TX --> SUM[Summarisation<br/>names forbidden,<br/>claimed labels become participant]
    SUM --> SP[(Postgres<br/>summaries & paragraphs)]
    SP --> TH[(Postgres topic history)]
    TH --> G[(Neo4j topic graph<br/>places, organisations,<br/>concepts, themes.<br/>Never people)]
    TX -.-> MX[Moderator analysis<br/>group patterns, no attribution<br/>without speaker labels]
    MX -.-> TR[(Extraction traces<br/>off in production,<br/>90 days)]

    classDef raw fill:#fde8e8,stroke:#c0392b,color:#000
    classDef clean fill:#e8f5e9,stroke:#2e7d32,color:#000
    class S3 raw
    class TX,SP,TH,G clean
```

Red holds identifiable material (the audio itself). Green has been through redaction. Redaction is automatic, not perfect, and uploaded documents are not redacted. See the warning on [Privacy](https://docs.kair.is/platform/privacy.md#built-for-the-group-not-the-individual).

***

## What we keep and what we never keep

| | Kept? | Detail |
| - | - | - |
| Original audio | **Kept** | Until the session is deleted or erased (Recall's copy of a meeting-bot recording is deleted with it). Encrypted at rest (SSE-C). Served only through the API with signed, expiring links. An S3 lifecycle rule (30 days in the [configuration example](https://docs.kair.is/platform/configuration.md#s3-lifecycle)) is optional and operator-set |
| Transcript text | **Kept, redacted** | Person names replaced before storage; contact identifiers removed by the regex redactor |
| Summaries, topics, topic graph | **Kept, anonymised** | About the group's discussion. People are never graph entities |
| Voiceprints, cross-session speaker identity | **Never stored** | Removed together with diarisation. There is nothing to extract or erase |
| Who said what | **Not recorded** | Only labels a source supplies (Recall, phone track names), or a label a member chose to claim. Without a claim nothing links a remark to a person |
| Uploaded documents | **Kept, not redacted** | Deleted on erasure when the uploader is known |
| Audit of API calls (`audit_events`) | **Kept, no content** | Actor, method, path, status, hashed IP. Append-only; rows can be deleted after two years. Survives erasure and is included in the actor's own export |
| Audit of row changes (`audit_trail`) | **Kept, personal fields stripped on erasure** | Tokens, secrets, hashes and free text are never written to it |
| Erasure bookkeeping (`gdpr_log`) | **Kept, no personal data** | Event kind and request id only |
| Moderator extraction traces | **Off in production & development, on in staging by default** (`moderator_agent.trace_extractions`) | Hold prompts and raw model output. Hard-deleted after `trace_retention_days` (default 90) |
| Sentry | **No personal data sent** | Users identified by id only |
| PostHog | **No page text or secrets** | Session replay masks all text & inputs; tokens, OAuth codes & URL fragments are stripped from captured URLs, nested redirect links included |
| Minimal-trace export | **Not stored** | Built on request, downloaded in the browser, nothing kept on the server |

### Short-lived data

| Data | Lifetime |
| - | - |
| Live recording status (Redis) | 1 hour TTL |
| Step-up code | 15 minutes, only a hash stored |
| Fresh sign-in required for erasure and every export | 10 minutes |
| Data request cancel link | Token stored as SHA-256 only; valid while the request is pending |
| Export download link (emailed) | 24 hours |
| Export archive in S3 | Deleted 3 days after it finished or failed |
| Erasure deadline | 30 days from the request, with admin reminders at 7 and 21 days and escalation at 28 |

***

## Personal versus organisation level

Every workspace belongs to an organisation, and the organisation's kind decides who controls its data.

```mermaid theme={null}
flowchart TB
    subgraph TEAM["Team organisation: organisation is controller, Kair is processor"]
        direction TB
        T1[Member requests erasure] --> T2{Org admin<br/>or platform admin}
        T2 -->|reject with reason| T3[Account unfrozen,<br/>subject emailed]
        T2 -->|approve| T4[Erase the member's content<br/>in the organisation's workspaces,<br/>redact them out of the rest]
        T4 --> T5{Last team<br/>membership?}
        T5 -->|no| T6[Account stays,<br/>other teams untouched]
        T5 -->|yes| T7[Also delete Personal workspaces,<br/>credentials, devices,<br/>pseudonymise user row]
    end
    subgraph PERSONAL["Personal organisation: Kair is controller"]
        direction TB
        P1[Owner requests erasure<br/>in Personal organisation] --> P2{Platform admin}
        P2 -->|approve| P3[Personal workspaces deleted outright,<br/>shared or not, plus account records]
    end
```

| | Team organisation | Personal organisation |
| - | - | - |
| Controller | The organisation | Kair |
| Who decides an erasure request | Org admins, or a platform admin | A platform admin |
| What an erasure removes | The person's content in the organisation's workspaces; their sessions are tombstoned and everything derived from them is cleaned | Their private workspaces, deleted outright even when others were invited |
| Other people's sessions | Kept as the organisation's record, the person's speaker claims removed | Not applicable, a Personal organisation has one member |
| Account-level data | Removed only when it is the person's last team | Removed with the request |
| While pending | Whole account frozen, API keys off, calendar auto-join paused | Same |

A member who leaves a workspace does not take their contributions with them. Their only route is a data request, and for a single remark in someone else's organisation, asking that organisation's admin. Leaving or being removed clears their ownership of sessions in that workspace (`sessions.owner_user_id`), so they lose access; the session keeps `former_owner_user_id`, which erasure and export still treat as theirs.

***

## What an erasure does to each kind of data

```mermaid theme={null}
flowchart LR
    ER{{Erasure approved}} --> DEL[Deleted]
    ER --> SCR[Scrubbed:<br/>text replaced by<br/>gdpr-scrubbed marker,<br/>row and counts kept]
    ER --> KEEP[Kept]

    DEL --> D1[Audio, chunks,<br/>Recall recordings]
    DEL --> D2[Transcript segments,<br/>notes, meeting chat,<br/>work items, project candidates,<br/>reports quoting their sessions]
    DEL --> D3[Documents the person uploaded;<br/>all audio & documents in<br/>their Personal workspaces]
    DEL --> D6[Their export archives]
    DEL --> D4[Memberships, shares, claims,<br/>calendar links, assistant chats]
    DEL --> D5[Topics only they held,<br/>appearances from their sessions]

    SCR --> S1[Transcript and summary text<br/>of their sessions]
    SCR --> S2[Session title and metadata<br/>tombstoned]
    SCR --> S3x[Moderator insights, critiques,<br/>feedback comments,<br/>extraction traces]
    SCR --> S4[audit_trail personal fields]

    KEEP --> K1[Other people's sessions,<br/>as the organisation's record]
    KEEP --> K2[Scores, ratings, counts<br/>for pattern analysis]
    KEEP --> K3[audit_events and gdpr_log,<br/>no personal content]
```

The neo4j topic graph is re-projected after topic history is cleaned, so it holds no erased session. The step-by-step table, including checkpointing and retries, is in [Data subject requests](https://docs.kair.is/platform/data-subject-requests#what-is-deleted-vs-retained).

***

## Exports by level

| | Personal | Workspace | Organisation (GDPR) | Minimal-trace |
| - | - | - | - | - |
| Requested by | Any member | Workspace owner | Org admin | Org admin |
| Org kind | Any | Any | Team | Team only |
| Includes audio | Yes, sessions you own | No | No | No |
| Includes transcripts and summaries | Yours, plus lines under labels you claimed | Whole workspace | Every workspace in the organisation | No |
| Includes members' private notes | Your own | No | No | No |
| Delivery | Emailed link, 24 hours | Emailed link, 24 hours | Emailed link, 24 hours | Direct download |
| Kept on server | 3 days | 3 days | 3 days | Nothing |

All four need a sign-in from the last 10 minutes, and API keys or OAuth/MCP tokens are refused. Full column-by-column detail: [Exports](https://docs.kair.is/platform/exports.md).

***

## Accountability trail

| Question | Where it is answered |
| - | - |
| Who called what, including admin reads | `audit_events` |
| Which rows changed | `audit_trail` (Postgres triggers, redacted on erasure) |
| Which erasures ran and when | `gdpr_log`, linked by data request id |
| Which MCP tools read content | `mcp_tool_calls`, see [MCP server](https://docs.kair.is/platform/mcp-server.md) |


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