Self-managed Redis Agent Memory prerequisites

Review software, Redis, network, Secret, image, and sizing prerequisites for self-managed Redis Agent Memory.

Agent Memory installs as the memory service of the redis-iris Helm chart 1.0.0, from the Redis AI Helm repository https://helm.redis.io/ai. The suite deploys the Agent Memory Data Plane, workers, and Control Plane, with the Identity Service and the Iris console. Start with the suite prerequisites. This page adds what Agent Memory needs.

What you need

Item Where it comes from
Helm chart redis-ai/redis-iris 1.0.0
Container images Docker Hub: redislabs/agent-memory:1.0.0 and redislabs/agent-memory-control-plane:1.0.0, plus the Identity Service and console images the suite installs
Redis databases You provide Store Redis, Job Redis, and Metadata Redis
License key One suite license. Contact your Redis representative or contact sales.
Provider credentials You provide embedding provider credentials and, when worker features are enabled, promotion or summarization LLM credentials

Redis databases

The suite chart does not deploy Redis databases. Provision them yourself and pass their URLs in the overlay Secret that memory.secrets.secretName names. The Data Plane and the Control Plane both read the overlay, so they use the same Metadata Redis and Store Redis.

Store Redis must support Search and JSON capabilities because Redis Agent Memory creates JSON and vector search indexes for memory data. Job Redis and Metadata Redis do not need those capabilities when they are deployed as separate Redis databases.

Store Redis must run Redis 7.2.0 through 8.4.x. Job Redis and Metadata Redis must run Redis 6.2 or later.

Redis database Required Key in the overlay Secret Purpose
Store Redis Always databases."1".urls Session memory JSON, long-term memory hashes, RediSearch indexes, vectors, and TTL-managed data.
Metadata Redis Always metadata.urls, and optionally metadata.namespace (default iris:memory) Store records.
Job Redis Always background_jobs.redis.urls and, when execution idempotency is enabled, background_jobs.idempotency.urls Background work, retry state, delayed jobs, and execution markers. The two settings can use the same Redis.

The Identity Service keeps agent-key records in its own Metadata Redis, which it reads from the Secret in identityService.metadata.existingSecret. The Data Plane checks agent keys by calling the Identity Service.

For a lab deployment, the Redis roles can point to the same Redis endpoint if it has the required modules and capacity. For production, separate Store Redis, Job Redis, and Metadata Redis when possible so memory data, background work, and control metadata can be scaled, backed up, and operated independently.

For Job Redis, use a non-evicting policy such as noeviction or volatile-ttl.

Metadata Redis durability

Metadata Redis is small compared with Store Redis, but it is operationally critical. Use persistent storage, Redis authentication, network isolation, and TLS where required.

Avoid eviction of metadata keys. Losing Metadata Redis removes Control Plane store records, and losing the Identity Service Metadata Redis removes agent-key records.

Network access

  • Connected install: the cluster must be able to pull images from docker.io and reach https://helm.redis.io/ai.
  • Air-gapped install: mirror the images into an internal registry, as described in Air-gapped and private registry installs.
  • Runtime access: Redis Agent Memory pods must reach the Redis databases and any embedding or LLM provider endpoints used by the deployment.
  • Identity Service: the Data Plane calls iris-identity-service:9200, and the Identity Service calls the Control Plane on port 9100 to validate grants. Allow both paths if you restrict traffic.
  • Data Plane exposure: use NetworkPolicy, ingress, gateway, service mesh, private load balancer, or equivalent controls to restrict API access.

Credentials and Secrets

Create these Secrets before you install. Deploy and Create the shared Secrets show each one.

Secret Values key Holds
Suite license global.iris.license.existingSecret The suite license key.
Overlay memory.secrets.secretName Redis URLs and provider credentials, in the key overlay.yaml.
Identity Service metadata identityService.metadata.existingSecret The Identity Service's Metadata Redis URL.
Cluster CA memory.tls.caCertSecret The cluster CA, so workers can verify the ServiceAccount issuer.

The suite generates Memory's runtime credential, Control Plane internal token, and admin token (iris-memory-admin-token). See Credentials.

Air-gapped and private registry installs

Mirror these Agent Memory images into your internal registry, along with the images of the other components the suite installs:

Image Needed
redislabs/agent-memory:1.0.0 Always (Data Plane and workers)
redislabs/agent-memory-control-plane:1.0.0 Always
redislabs/iris-identity-service:1.0.0 Always
curlimages/curl:8.17.0 The suite preflight (preflight.image)
python:3.13-alpine helm test iris (tests.suiteVerify.image)
hibiken/asynqmon:0.7.2 Optional, with memory.controlplane.queueMonitor
docker pull redislabs/agent-memory:1.0.0
docker tag redislabs/agent-memory:1.0.0 \
  registry.example.com/redislabs/agent-memory:1.0.0
docker push registry.example.com/redislabs/agent-memory:1.0.0

Repeat for each image. If the registry requires authentication, create an image pull Secret:

kubectl -n <namespace-name> create secret docker-registry iris-registry \
  --docker-server=registry.example.com \
  --docker-username=<username> \
  --docker-password=<password>

Add the registry settings to iris-values.yaml:

global:
  imagePullSecrets:
    - name: iris-registry
identityService:
  image:
    repository: registry.example.com/redislabs/iris-identity-service
memory:
  image:
    repository: registry.example.com/redislabs/agent-memory
  controlplane:
    image:
      repository: registry.example.com/redislabs/agent-memory-control-plane
  airgap:
    enabled: true

memory.airgap.enabled: true checks Memory's own image repositories only. Confirm that every configured repository is reachable from the cluster. Omit the pull Secret if the registry doesn't require authentication.

System requirements

The suite's replica defaults are in Availability. During a rolling update, Kubernetes runs old and new pods at the same time, so a small two-node test cluster can run out of CPU during install or upgrade. Use at least three nodes, or enough CPU headroom for the rolling-update overlap.

For a constrained lab cluster, reduce Memory's replicas explicitly:

memory:
  server:
    replicaCount: 1
    autoscaling:
      enabled: false
  worker:
    replicaCount: 1
    autoscaling:
      enabled: false

Do not use reduced replica counts as the production HA recommendation.

Helm values to review

These keys go under memory: in iris-values.yaml. The values the suite sets itself are listed in Add the Agent Memory values.

Area Values Use when
Image image.repository, image.tag Selecting a private registry or mirrored image.
Air-gapped installs airgap.enabled Validating a disconnected or private-registry install.
Chart-rendered configuration shared, memory, controlplane.configData Setting the Data Plane and Control Plane configuration from Helm values.
Credentials overlay secrets.* Naming the overlay Secret and its key, or layering more overlay Secrets.
API server capacity server.resources, server.autoscaling.* Tuning request capacity or memory footprint.
Worker capacity worker.resources, worker.autoscaling.* Tuning background job throughput.
Scheduling server.nodeSelector, worker.nodeSelector, server.affinity, worker.affinity, server.tolerations, worker.tolerations Controlling pod placement.
Networking service.type, ingress.* Exposing Redis Agent Memory outside the cluster. The Ingress sets no controller annotations by default and routes paths without rewriting, so it works on any controller, such as ingress-nginx or Traefik. Set ingress.className (default nginx) to your controller's class.
Service account serviceAccount.* Matching customer namespace security policy.
Private certificate authority (CA) tls.caCertSecret Trusting a private CA for outbound connections, such as the cluster CA for worker identity.
Control Plane controlplane.image.*, controlplane.configData, controlplane.adminToken.* Mirroring the Control Plane image, setting its configuration, or bringing your own admin token.

Do not use floating image tags in production.

RATE THIS PAGE
Back to top ↑