Add self-managed Agent Memory docs

[n-valchev]

Summary

• add the self-managed Redis Agent Memory documentation hierarchy
• add static stores and Control Plane managed stores walkthroughs
• add shared Data Plane configuration, authentication, operations, troubleshooting, API examples, and Control Plane API reference
• link the self-managed path from the Agent Memory overview

Validation

• git diff --check
• git diff --cached --check
• jq empty content/develop/ai/context-engine/agent-memory/self-managed/control-plane-api-reference/openapi-control-plane.json
• /Users/nikolay.valchev/go/bin/hugo --destination /tmp/redis-docs-hugo-check

---

Note

Low Risk
Documentation-only changes with no application code; security guidance is descriptive for operators.

Overview
Adds a self-managed Agent Memory doc tree for private-preview Kubernetes deployments, covering planning, prerequisites, Data Plane config, Helm walkthroughs for static stores vs Control Plane–managed stores, authentication (admin token, agent keys, worker identity, gateways), curl API examples, operations, and troubleshooting.

The Agent Memory overview Redis Software tab now links to this section and notes the license requirement. The shared Data Plane API page is retitled Redis Agent Memory API (from LangCache), and the OpenAPI spec title/description are generalized for cloud and self-managed auth. A new Control Plane API reference page ships with an OpenAPI document for store and agent-key admin endpoints.

^{Reviewed by Cursor Bugbot for commit 3ec1dd3. Bugbot is set up for automated code reviews on this repo. Configure here.}

[mich-elle-luna]

Hi @n-valchev thanks for this PR, here is the feedback on technical accuracy, otherwise it looks good to me:

Issue 1 — Control Plane embedding config is incomplete (worth clarifying)
In deploy-control-plane.md, the sample controlplane-onprem.config.yaml only specifies:

embedding:
dimensions: 3072
The Data Plane config (and the authentication.md agent-key example) both use the full shape:

embedding:
provider: openai
models:
default_embedding_model: text-embedding-3-large
dimensions: 3072
This may be intentional — the Control Plane might only need dimensions to create indexes and doesn't do embedding itself. But it's not explained, and a reader comparing the two configs will be confused. Does the Control Plane need provider and models, or is dimensions sufficient? If dimensions-only is correct, add a brief comment explaining why.

Issue 2 — shasum produces SHA-1, not SHA-256
Both deploy guides use:

LICENSE_CHECKSUM="$(shasum ./license | awk '{print $1}')"
shasum without flags defaults to SHA-1 on macOS and Linux. For a change-detection checksum this works fine, but it looks like a weak hash to security-conscious readers. More importantly, on some Linux systems shasum isn't installed by default (it comes from perl), while sha256sum is universally available.

Recommendation: use sha256sum ./license | awk '{print $1}' or shasum -a 256 ./license | awk '{print $1}'.

Issue 3 — helm repo add --force-update requires Helm 3.7.2+
Both deploy guides use:

helm repo add redis-ai https://helm.redis.io/ai --force-update
The --force-update flag was introduced in Helm 3.7.2. The prerequisites page says Helm 3.x (minimum 3.x). If someone has Helm 3.0–3.7.1, this command silently fails or errors depending on version.

Recommendation: Either raise the Helm minimum version in prerequisites to 3.7.2, or drop --force-update (it's not required — helm repo update redis-ai handles the refresh).

[n-valchev]

Thanks @mich-elle-luna. I've addressed the comments.

[mich-elle-luna]

thank you!