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

# Security

> How Foggy protects your data — read-only access, encryption, and data isolation.

Every non-Slack integration is read-only — Foggy delivers findings, not actions. Credentials are encrypted at rest and never sent to LLM providers. Investigation data is never used to train models.

## Read-only access

Foggy runs with read-only permissions for observability and infrastructure integrations — no agents, no sidecars, no writes to production. All non-Slack integrations are enforced as read-only at the connector level, regardless of the API token permissions you grant.

The exception is Slack, where Foggy sends investigation results and alert notifications.

## Credential encryption

All connector credentials (API tokens, URLs) are encrypted at rest using **Fernet symmetric encryption**:

* Credentials are encrypted the moment you save them
* Decrypted only when making API calls to your tools
* Never logged, cached, or exposed in the UI after initial entry
* Never sent to LLM providers — only query results are sent for analysis

## Project isolation

All data is scoped to **projects** (workspaces). Users can only access projects they belong to.

| Role       | Permissions                                                                              |
| ---------- | ---------------------------------------------------------------------------------------- |
| **Owner**  | Full control — settings, connectors, knowledge base, automations, member management      |
| **Member** | Chat and view — can investigate and view results, but can't change project configuration |

See [Teams and access](/platform/teams-and-access) for invite and workspace-setup details.

## Integration scopes

Foggy requests the narrowest scopes each integration supports. Setup details are in each integration's page under **Integrations**.

| Integration   | Access                                                                                        | Direction                                                   |
| ------------- | --------------------------------------------------------------------------------------------- | ----------------------------------------------------------- |
| Grafana       | Read-only Viewer token (optional for public dashboards)                                       | Query only                                                  |
| Prometheus    | HTTP read; bearer token optional                                                              | Query only                                                  |
| Kubernetes    | Read-only `ClusterRole` (self-hosted in-cluster) or via Satellite for any other cluster       | Query only                                                  |
| OpenSearch    | Read-only bearer token                                                                        | Query only                                                  |
| GitHub        | Fine-grained PAT: read on Contents, Actions, Issues, Pull requests                            | Query only (writes require explicit opt-in)                 |
| Sentry        | Personal token with `org:read`, `project:read`, `team:read`, `event:read`, `alerts:read`      | Query only                                                  |
| Datadog       | API key + Application key with `mcp_read` and read on Monitors, Logs, APM, Metrics, Incidents | Query only                                                  |
| Linear        | Restricted (read-only) API key; full-access only when you want Foggy to file tickets          | Query only by default                                       |
| Slack         | OAuth app with `chat:write` plus channel and mention read scopes                              | **Bidirectional** — Foggy reads mentions and writes replies |
| Alert Webhook | Inbound HTTPS with a Bearer token per workspace                                               | Inbound only                                                |
| Satellite     | Outbound-only TLS from your cluster with ingest token; credentials stay local                 | Outbound from your network                                  |

## Data handling

* **No model training** — Investigation data is never used to train models or shared across projects.
* **Encryption in transit and at rest** — Connector credentials use Fernet symmetric encryption.
* **Credential isolation from LLMs** — API tokens and connection strings are never sent to LLM providers. Only query results (metrics, logs, events) are sent for analysis.

## Bring Your Own Model

Use your own LLM API keys to control which provider processes your data. Choose the model per-message via the model selector in the chat input.

## LLM data flow

What data goes where:

| Data type                 | Sent to LLM? | Details                                                                          |
| ------------------------- | :----------: | -------------------------------------------------------------------------------- |
| Investigation prompts     |      Yes     | Your question is sent to the selected LLM for reasoning                          |
| Tool query results        |      Yes     | Metrics, logs, and events from your tools are sent for analysis                  |
| API tokens / credentials  |   **Never**  | Credentials are used locally to query your tools, never included in LLM requests |
| Connection strings / URLs |   **Never**  | Used locally for connector API calls only                                        |
| Knowledge Base entries    |      Yes     | Enabled entries are included as context for investigations                       |

<Info>
  Questions about security? [Contact us](https://foggyhq.com) — we're happy to discuss your specific compliance requirements.
</Info>