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

# Pavo Cloud (SaaS)

Pavo Cloud is the managed deployment: Pavo hosts and operates the entire platform in its managed cloud (GCP, on GKE). There is no infrastructure for your team to stand up or run — the deployment is live the week you connect your sources.

<Note>
  Your team's only setup work is pointing Pavo, read-only, at your sources and configuring SSO — roughly half a day of infra effort, all front-loaded in the first week.
</Note>

## Tenancy and isolation

Pavo Cloud is a **multi-tenant** platform with per-organization isolation. Each organization runs in its own isolated namespace, with dedicated data stores, search indices, and endpoints. Your organization's data is never shared with, or visible to, another tenant.

* Encrypted at rest (AES-256 via cloud KMS) and in transit (TLS 1.2+); customer-managed encryption keys (CMEK) available on request.
* No public data endpoints; network isolation throughout.
* Sign-in runs through your identity provider via SSO (OIDC).

## Architecture

```mermaid theme={null}
flowchart TB
  subgraph sources["Your cloud / SaaS — read-only sources"]
    direction LR
    repos["Code<br/>repositories"]
    wh["Data<br/>warehouse"]
    exp["Experimentation<br/>platform"]
    dash["Dashboards"]
    docs["Documentation"]
  end

  sources -->|"Read-only ingest — connectors pull, nothing writes back"| app

  subgraph cloud["Pavo Cloud — multi-tenant · per-organization isolation"]
    app["Pavo application<br/>agents · knowledge pipeline · connectors · sandbox"]
    obj["Object storage<br/>raw artifacts · backups"]
    pg["Postgres<br/>app state · metadata"]
    rq["Redis + queues<br/>cache · async work"]
    idp["Zitadel<br/>SSO via your IdP"]
    kms["KMS + secrets<br/>never indexed"]
    app --- obj
    app --- pg
    app --- rq
  end

  app -->|"Under DPA"| msp["Managed sub-processors<br/>Elasticsearch · Temporal · Grafana<br/>Amplitude · Brevo"]
  app -->|"Zero data retention — no training on your data"| llm["Model providers<br/>OpenAI · Anthropic"]
```

*Figure — Pavo Cloud: multi-tenant platform with per-organization isolation; managed sub-processors under DPA.*

The sandbox runs agent-generated code behind a **default-deny egress allowlist proxy** — outbound access is limited to an agreed allowlist (open-source package registries by default), and everything else is blocked. See [Security → Network & egress](/security/overview#network--egress).

## Components

| Component             | Purpose                                                                             | Notes                                                       |
| :-------------------- | :---------------------------------------------------------------------------------- | :---------------------------------------------------------- |
| Application + sandbox | Agents, knowledge pipeline, connectors; sandbox executes agent code                 | Runs on Pavo's GKE; egress via default-deny allowlist proxy |
| Object storage        | Raw ingested artifacts and index backups                                            | GCS, isolated per organization                              |
| Database              | Application state, connector config, job metadata                                   | Cloud SQL (Postgres)                                        |
| Cache & queues        | Caching, rate limiting, async work queues                                           | Redis + Pub/Sub; transient operational data                 |
| Search index          | Search & retrieval over processed representations                                   | Elasticsearch (managed), per-org isolated                   |
| Workflow engine       | Durable execution of agent workflows                                                | Temporal (managed), mTLS                                    |
| Observability         | System health metrics and dashboards — operational telemetry only, no customer data | Prometheus + Grafana (managed)                              |
| Model inference       | LLM reasoning and embeddings; no training on your data                              | OpenAI / Anthropic under zero-data-retention agreements     |
| Identity & SSO        | Authentication via OIDC, integrating your identity provider                         | Zitadel; platform user identities only                      |
| Secrets & keys        | Credentials and encryption keys — never written to any index                        | Secret Manager + Cloud KMS                                  |

## Sub-processors in this deployment

Pavo Cloud uses a defined set of managed services that process data on Pavo's behalf, each disclosed and under a DPA: **Elasticsearch**, **Temporal**, and **Grafana** (core platform services), **Zitadel**, **Amplitude**, and **Brevo** (supporting services — no customer business data), and **OpenAI** / **Anthropic** for inference under zero-data-retention agreements. Optional integrations (**Modal**, **Langfuse**, **Parallel**) are individually gated per instance and disabled unless you enable them.

See the full list, with the data each one processes, in [Sub-processors](/security/sub-processors).

## What you provide

* Read-only credentials for the sources you connect (warehouse, code repositories, experimentation platform, dashboards, documentation).
* SSO configuration through your identity provider (OIDC).

That's the entire footprint — there is nothing to provision, patch, or scale on your side.

## Next steps

<CardGroup cols={2}>
  <Card title="Customer VPC (BYOC)" icon="server" href="/deployment/customer-vpc">
    Need the platform inside your own cloud boundary? Run it in your VPC.
  </Card>

  <Card title="Security" icon="shield" href="/security/overview">
    Certifications, data handling, controls, and the full sub-processor list.
  </Card>
</CardGroup>