---
title: "Choose Where a proxymock Replay Runs | Speedscale Docs"
description: "proxymock can replay recorded traffic in three places: in-process on your machine, inside your Kubernetes cluster with no Speedscale cloud involved, or through Speedscale cloud like a dashboard replay. Compare the three and run each one from proxymock web, the CLI, or the MCP server."
canonical: "https://docs.speedscale.com/proxymock/guides/replay-paths/"
---

[Documentation index](https://docs.speedscale.com/proxymock/llms.txt) | [Markdown](https://docs.speedscale.com/proxymock/guides/replay-paths/index.md) | [Canonical page](https://docs.speedscale.com/proxymock/guides/replay-paths/)

# Choose Where a proxymock Replay Runs

Every replay sends recorded requests at a target and reports how the responses compare. What changes is _where_ the replay runs and _what it needs from you_. proxymock offers three paths, and they are independent: having a cluster does not stop you replaying locally, and being logged in to Speedscale cloud does not force a replay through it.

|  | On this machine | In your cluster | Via Speedscale cloud |
| --- | --- | --- | --- |
| Where traffic is sent from | proxymock, in-process | a generator pod the operator creates | a generator pod the operator creates |
| Target | any URL your machine can reach | a workload or Service in the cluster | a workload in a cluster registered with your tenant |
| Needs a kubeconfig | no | yes | no |
| Needs the Speedscale operator | no | yes | yes |
| Needs a Speedscale cloud login | no | no | yes |
| Does anything leave the cluster or laptop | no | no | yes: the recordings are pushed as a snapshot |
| Where the report lands | your workspace | the cluster, read with proxymock | the Speedscale dashboard |

In proxymock web all three live in the **Where it runs** strip at the top of the Replay tab. Picking one decides what the rest of the form asks for, and the run button changes with it. The **Test config** picker below it applies to all three paths; see [Configure a Replay with Test Configs](https://docs.speedscale.com/proxymock/guides/test-configs/).

<a id="on-this-machine"></a>

## On this machine

If your traffic is already in a Speedscale Cloud snapshot, [pull it and replay it locally](https://docs.speedscale.com/proxymock/guides/cloud-snapshot-local/). That guide covers cloud authentication, the download, dependency mocks, and the local target.

The replay runs inside proxymock and sends traffic to a URL. Use it while you are editing code: your service runs locally, the replay hits it directly, and nothing else has to exist.

![The Replay tab with On this machine selected and a local URL target](https://docs.speedscale.com/assets/images/replay-where-local-eecd0033dae9ee8718750d70eb68f833.png)

1.  Start your service so it is listening.
2.  In the Replay tab pick **On this machine**.
3.  Select the recordings to replay.
4.  Leave routing on **All traffic to one destination** and set the URL your service listens on.
5.  Press **Run replay**.

The same run from the CLI:

```shell
proxymock replay --in ./proxymock --test-against localhost:3000
```

If your service calls dependencies you do not want to reach for real, run `proxymock mock` alongside it and point the service at the mock. See [Mock a dependency](https://docs.speedscale.com/proxymock/guides/).

A replay that cannot open a connection fails with `tcp probe against "HOST:PORT": unable to connect`. That means nothing was listening, so start the service first.

<a id="in-your-cluster"></a>

## In your cluster

The replay runs inside Kubernetes against a real workload, and **no part of it involves Speedscale cloud**. proxymock analyzes the recordings locally, stages the snapshot in the in-cluster forwarder over a port-forward, and the operator runs the replay from there. The report stays in the cluster.

![The Replay tab with In your cluster selected, showing the cluster context and namespace pickers](https://docs.speedscale.com/assets/images/replay-where-cluster-c2c2df8b0faba1eee1d6f246c91bb552.png)

Prerequisites: a kubeconfig context for the cluster and the Speedscale operator installed in it. No login.

1.  Start proxymock web with the context you want, for example `proxymock web --kube-context minikube`.
2.  Pick **In your cluster**, then the cluster context and the namespace your workload runs in.
3.  Set the destination to a **Workload** or **Service**. A plain URL is resolved inside the generator pod, where your laptop's `localhost` does not exist.
4.  Every recorded dependency is mocked by default, so the workload answers from the recording instead of reaching the real thing. To mock only some, or to let the workload reach its real dependencies, open **Options** and change **Mock dependencies**.
5.  Press **Run in cluster**.

The same run from the CLI:

```shell
proxymock cluster replay start --in ./proxymock -n my-namespace --workload my-service --wait -o json
```

The result reports `snapshotSource`. `local` means the snapshot was staged in the cluster. `cloud` means staging failed and proxymock fell back to pushing it, which needs a login. Watch and read a run with `proxymock cluster replay status` and `proxymock cluster replay logs`; see [Work a Kubernetes Cluster from the CLI](https://docs.speedscale.com/proxymock/guides/cluster/).

<a id="via-speedscale-cloud"></a>

## Via Speedscale cloud

This is the same replay the Speedscale dashboard starts. The recordings are pushed as a snapshot, Speedscale cloud starts the replay in the cluster you choose, and the report lands in the dashboard where it can be shared, compared, and kept.

![The Replay tab with Via Speedscale cloud selected, showing the cluster and test config pickers](https://docs.speedscale.com/assets/images/replay-where-cloud-4dfb364923f12e3975a736c38bdc38fd.png)

Prerequisites: a Speedscale cloud login (`proxymock init`) and a cluster registered with your tenant, which means the operator there is connected to Speedscale cloud.

1.  Pick **Via Speedscale cloud**.
2.  Choose the **Cluster**, the **Namespace**, and the workload under test. These lists come from your tenant, not your kubeconfig, so you do not need cluster credentials on this machine.
3.  Choose a **Test config** if you want something other than `regression`.
4.  Press **Push and run via cloud**.

The same run from the CLI:

```shell
proxymock cloud replay --in ./proxymock --cluster my-cluster -n my-namespace --workload my-service --wait
```

Add `--dry-run` to print what would be pushed and started without doing either. `--wait` polls until the report reaches a terminal status and prints its dashboard link. Like the in-cluster path, a workload replay mocks every recorded dependency unless you narrow it with `--mock` or turn it off with `--no-mocks`. This is the proxymock equivalent of [replaying with speedctl](https://docs.speedscale.com/guides/replay/via-speedctl/).

<a id="from-an-ai-agent"></a>

## From an AI agent

The MCP server exposes all three paths, so an assistant can pick one:

| Path | Tool and action |
| --- | --- |
| On this machine | `replay_traffic`, with `mock_server_start` for dependencies |
| In your cluster | the `cluster` tool, action `replay-start`, then `replay-status` and `replay-logs` |
| Via Speedscale cloud | the `cloud_replay` tool: `clusters`, `namespaces`, `workloads`, `start`, `status` |

`replay-start` takes `snapshot_source`. The default, `auto`, stages the snapshot in the cluster and falls back to a cloud push. Use `local` to refuse the fallback, so a run that cannot stage in the cluster fails instead of sending recordings to Speedscale cloud. See the [MCP tools reference](https://docs.speedscale.com/proxymock/how-it-works/mcp-tools/).
