---
title: Local, cloud, or self-hosted
description: The three ways to run ClawMetry — fully local, local compute with an encrypted cloud display, or a self-hosted ingest — and what each implies for your data.
keywords: ClawMetry self-hosted, local agent observability, air-gapped agent monitoring
eyebrow: Get started
---

# Local, cloud, or self-hosted

The compute always happens on your machine. What differs is where the *display*
lives and who stores the snapshot.

## Local only

Nothing leaves the machine. No account, no key, no network call.

```bash
clawmetry onboard --local
```

This writes a persistent local-only marker. A later `clawmetry connect` will
refuse to override it unless you pass `--force`, which exists so that an
automated provisioning script cannot quietly opt a machine into cloud sync.

**Choose this when** you are a single developer on one machine, or when policy
forbids agent data leaving the host at all.

**What you give up:** the cloud dashboard, multi-node fleet view, phone
approvals, and the device display. Everything else — every runtime, every
dashboard page, Guard, detectors, the MCP server, the query API — works
identically.

## Local compute, encrypted cloud display

The default when you say yes to an account. The daemon still does all the
reading and computing locally; it additionally pushes an encrypted snapshot so
you can open the dashboard from another device.

```bash
clawmetry connect --key cm_xxx
```

Session content is encrypted with AES-256-GCM before upload using a key held
only on your machine. The server stores ciphertext it cannot read and your
browser decrypts it. Aggregate counters travel in plaintext so a summary and a
device display can render without your key. The exact split is in
[End-to-end encryption](/docs/cloud/encryption/).

**Choose this when** you run agents on more than one machine, want alerts that
reach your phone, or want to look at a fleet without SSHing into each node.

Pause or stop it at any time:

```bash
clawmetry disconnect    # stops sync and removes the key
```

## Self-hosted ingest

You run the ingest endpoint. Nodes push to your server instead of
`ingest.clawmetry.com`, and you own the storage.

```bash
CLAWMETRY_ENDPOINT=https://clawmetry.internal.example.com clawmetry connect --key ...
```

The self-hosted server exposes the same ingest surface — heartbeat, events,
sessions, logs, memory, system snapshots, approvals — plus a read API for the
dashboards you point at it. Relevant environment: `CLAWMETRY_SELF_HOSTED`,
`CLAWMETRY_SELF_HOSTED_DB`, `CLAWMETRY_SELF_HOSTED_E2E`.

**Choose this when** you need data residency guarantees, an air-gapped network,
or an audit trail inside your own perimeter. See
[Self-hosted ingest](/docs/cloud/self-hosted/) and
[Plans and entitlements](/docs/ops/plans/) for the licensing side.

## Comparison

| | Local only | Cloud display | Self-hosted |
|---|---|---|---|
| Account required | No | Yes | Yes (license) |
| Data leaves the host | Never | Encrypted snapshot | To your server |
| Server can read content | — | No | Your policy |
| Multi-node fleet | No | Yes | Yes |
| Phone approvals / device | No | Yes | Yes |
| All {{RUNTIME_COUNT}} runtimes | Yes | Yes | Yes |
| Guard and detectors | Yes | Yes | Yes |
| MCP server | Yes | Yes | Yes |

## Switching later

Nothing here is a one-way door. The store is the same file in all three modes,
so connecting or disconnecting changes only where snapshots go — your history
stays.

```bash
clawmetry status --json | jq '{cloud: .cloud, node_id: .node_id, endpoint: .endpoint}'
```