---
title: Add a Redis cache to your project
description: Create a private Redis cache on Google Memorystore (S to XL, optional HA and TLS) and connect it to a project as REDIS_URL, from the console or an AI tool.
url: https://agentlab.cresclab.com/docs/en/deploy/caches
language: en
updated: 2026-10-28
---

# Add a Redis cache to your project

Create a private Redis cache on Google Memorystore (S to XL, optional HA and TLS) and connect it to a project as REDIS_URL, from the console or an AI tool.

## The quickest way: let the deploy create it

If your code reads `REDIS_URL` (or another Redis connection variable) and the project has no value for it yet, deploying from an AI tool or uploading in the console creates a size S, Basic cache and connects it before the build starts. Creating it takes about 5 minutes. When this happens and when it doesn't: [Databases and caches the code needs](https://agentlab.cresclab.com/docs/en/deploy/mcp-tools.md#auto-provision).

## What you get

A dedicated Redis 7.2 instance on Google Memorystore, in Taiwan (`asia-east1`), for one environment.

| Setting             | Choices                                                                                                                                                                              |
| ------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| **Memory**          | S 1 GiB, M 2 GiB, L 4 GiB, XL 8 GiB. It can be expanded later, never reduced.                                                                                                        |
| **Availability**    | **Basic**, or **High availability** with a replica that takes over. Fixed once created.                                                                                              |
| **TLS encryption**  | Off by default. Fixed once created. With TLS, the project also needs the CA certificate.                                                                                             |
| **Eviction policy** | What happens when memory is full: `noeviction`, `allkeys-lru`, `allkeys-lfu`, `allkeys-random`, `volatile-lru`, `volatile-lfu`, `volatile-random` or `volatile-ttl`. Can be changed. |

- **Private.** The cache has a private address only. Only projects you connect can reach it, and only with the AUTH password, which is always on. You can't reach it from your own computer; from an AI tool, `caches_command` runs single commands for you.
- **What the plan allows.** Free has no caches. Pro allows 5, Team 20, Business by contract, all four sizes.
- **Price.** Cache prices aren't published yet: the console shows no estimate, and `caches_options` answers none. High availability costs more than Basic.

## Create a cache

**Console**

1. Open **Caches** and select **Create cache**.
2. Enter a **Name**: lowercase letters, digits and hyphens, starting with a letter, up to 63 characters.
3. Choose the **Memory**, **Availability**, **TLS encryption** and **Eviction policy**. For sessions or data you can't lose, use `noeviction`; for a pure cache, `allkeys-lru`.
4. Confirm. The cache is ready in about 5 minutes.

**AI tool**

```prompt
Create an Agent Lab Redis cache named sessions, size S, Basic, without TLS, eviction noeviction.
```

The assistant calls `caches_create`, shows you a preview, and creates it after you agree.

## Connect a project

Connecting writes the connection into the project's secrets and lets the project through the cache's network. Nobody sees the password.

**Console**

1. Open the cache, go to **Project connections** and select **Connect project**.
2. Choose the project and the variable names, then **Preview ENV changes**. The preview lists any variables that would be replaced.
3. Confirm, and keep **Restart project after injection** on so the project reads the new values.

**AI tool**

```prompt
Connect my orders-api project to the sessions cache as REDIS_URL.
```

The assistant calls `caches_binding_preview`, then `caches_binding_connect` with the preview's `token`. `names` maps each variable to a part of the connection:

```json
{ "REDIS_URL": "url", "REDIS_CA": "ca" }
```

Parts are `url`, `host`, `port`, `password` and `ca`. A cache with TLS needs `ca`.

The `url` part is `redis://:<password>@<private address>:6379/0`; with TLS it starts with `rediss://` and uses the port Memorystore gives a TLS cache (6378). Most clients take it as it is. With TLS, also give the client the CA certificate:

```js server.js
import { createClient } from "redis";

const redis = createClient({
  url: process.env.REDIS_URL,
  socket: process.env.REDIS_CA ? { tls: true, ca: process.env.REDIS_CA } : undefined,
});
await redis.connect();
```

A cache must be **Ready** before a project can connect. **Disconnect** removes the project's access; its credentials are removed once its network access is revoked.

## Change or delete a cache

- **Rename, expand memory or change the eviction policy** under **Manage**, or with `caches_update`. TLS and availability can't change: create a new cache instead.
- **Delete cache** deletes every key permanently. Project connections are revoked first.
- The cache's **Monitoring** shows memory, connections, commands, hit rate and evicted keys. Any evicted keys mean the cache is full: a larger size helps.

From an AI tool, `caches_update` and `caches_delete` need the cache's current `generation` from `caches_get`. `caches_command` runs one Redis command from an allowlist: reads at once, writes after you confirm. `AUTH`, `CONFIG`, `EVAL`, `FLUSH` and `KEYS` are refused. See [MCP tools › Redis (caches)](https://agentlab.cresclab.com/docs/en/deploy/mcp-tools.md#caches).