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

# Databricks data handling and privacy

> What leaves Databricks, what is stored where, audit logging, and the controls you have.

## What leaves Databricks

Requests to the Kyberis API contain exactly:

* **The indicators and queries you enrich** — IOC values (IPs, domains, URLs,
  hashes, emails), CVE IDs, actor, malware, or campaign names, or intel search
  text you type or select.
* **Environment context you explicitly provide** — for example the industry,
  products, and geography widgets in the CVE prioritization notebook.
* **`agent_context` metadata** — objective and requested-outcome strings, which
  are static defaults unless you override them, a workflow stage, and random run
  and step IDs. Nothing here is derived from your table contents.
* **Credentials** — the API key on the token-mint call only, and short-lived
  bearer tokens on enrichment calls.

Not sent, by design: table names, schemas, row contents beyond the selected
indicator column, cluster or workspace identifiers, user identities, and notebook
code.

## What is stored where

| Data | Location | Notes |
| - | - | - |
| API key | Databricks secret scope and app secret resources only | Never in code, widgets, job parameters, or files. See [Credential setup](/integrations/databricks/credentials). |
| Bearer tokens | App or notebook process memory | 30-minute TTL at most, never persisted |
| Enrichment results | Wherever you write them — display only by default, or a Delta table if you set `output_table` | Treat as security telemetry. See [Permissions](/integrations/databricks/permissions). |
| Kyberis-side processing | The Kyberis platform | Governed by your Kyberis subscription terms; queries are processed to serve the response, and product telemetry is retained per contract |

## Logging and audit

* **Databricks side.** Secret reads via `dbutils.secrets.get` are redacted in
  notebook output, and secret-scope access and app lifecycle events appear in
  Databricks audit logs under the `secrets` and `apps` event categories, so key
  access is attributable per user or principal.
* **Kyberis side.** Every API call is logged against the API key's principal with
  the `agent_context` run and step IDs. Use a distinct API key per workspace, or
  per team, so Kyberis-side usage and audit trails attribute cleanly, and pass a
  meaningful `run_id` — the helpers generate one per job run — to correlate a
  Databricks job run with Kyberis-side logs.
* **Error paths.** Exception messages produced by this integration never include
  credential material, and the `KyberisCredentials` type redacts the secret from
  `repr()`.

## Controls and opt-outs

* Enrich only the columns you choose; the helpers never scan tables themselves.
* Set `objective` widgets and inputs to neutral text if your change management
  requires it. They are free text and default to static strings.
* To keep results out of persistent storage, leave `output_table` empty, which is
  display-only, and export nothing.