MemoryStore API

This page lists every endpoint on the MemoryStore service (memorystore-api), with request fields, response shapes, and the exact error messages the server returns. For a guided introduction, start with the MemoryStore overview and quickstart.

Base URL

MemoryStore has its own API service, separate from the core platform API. In-cluster it is memorystore-api.cai-system.svc:8080.

No public API hostname yet

The platform is in alpha and does not publish a public hostname for this API. Reach it through a port-forward (shown below), or through an endpoint your administrator provides. The platformctl CLI does this automatically when $CAI_MEMORYSTORE_API is not set.

kubectl -n cai-system port-forward svc/memorystore-api 18081:8080 &
export CAI_MEMORYSTORE_API=http://localhost:18081

Authentication

Every route except GET /healthz and GET /livez requires a bearer credential:

Authorization: Bearer <token-or-api-key>

The credential is a session token (from POST /v1/auth/login on the core API, 12-hour life), an API key (prefix cai_), or the platform automation token. Roles are re-read from the database on every request. See API authentication.

An unauthenticated call gets 401 with this exact message: authentication required: sign in via agent-engine-api (POST /v1/auth/login) and send 'Authorization: Bearer <token>'.

Role	Can call
member	list, get, stats, create
admin	everything a member can, plus delete and rotate-credential

Conventions

• Error envelope. Every non-2xx response is {"error": "<message>", "request_id": "<id>"}.
• The 404 rule. "Not found" and "not yours" return byte-identical 404s. 403 appears only when you are a member but the action needs admin.
• Pagination. page_size (1–200, default 50) and page_token, with next_page_token in responses, same as every platform list.
• Request timeout. 30 seconds per request by default (REQUEST_TIMEOUT_SECONDS).
• Passwords are structural secrets. The instance object cannot contain a password — it has no field for one. The password appears in exactly two responses: create and rotate-credential.

Endpoints at a glance

Method	Path	Auth	Purpose
GET	/v1/projects/{projectID}/memorystores	member	List instances
GET	/v1/projects/{projectID}/memorystores/{name}	member	Get one instance
GET	/v1/projects/{projectID}/memorystores/{name}/stats	member	Live server statistics
POST	/v1/projects/{projectID}/memorystores	member	Create an instance
DELETE	/v1/projects/{projectID}/memorystores/{name}	admin	Delete an instance
POST	/v1/projects/{projectID}/memorystores/{name}/rotate-credential	admin	Mint a new password
GET	/healthz	none	Readiness check
GET	/livez	none	Liveness check

The instance object

Every read returns this shape:

{
  "name": "sessions",
  "resource_path": "projects/<short>/memorystores/sessions",
  "project_id": "<uuid>",
  "size_class": "medium",
  "version": "9.1.0-alpine3.23",
  "maxmemory_policy": "noeviction",
  "persistence": {"enabled": true, "size": "2Gi"},
  "state": "Ready",
  "ready": true,
  "external_exposure": false,
  "externally_reachable": false,
  "external_endpoint": "rediss://sessions-<short>.<domain>:443",
  "reachable_scope": "vpc",
  "memory_limit": "1Gi",
  "maxmemory": "768Mi",
  "ready_replicas": 1,
  "replicas": 1,
  "connection": {
    "host": "ms-sessions.cai-p-<short>.svc.cluster.local",
    "port": 6379,
    "username": "default",
    "credential_secret": "ms-sessions-credential"
  },
  "conditions": [{"type": "Ready", "status": "True", "reason": "...", "message": "..."}],
  "created_at": "2026-08-06T12:00:00Z"
}

Field	Type	Meaning
name	string	The instance name you chose. Immutable
resource_path	string	Stable path: projects/<short>/memorystores/<name>
project_id	string	The owning project's UUID
size_class	string	small, medium, or large
version	string	Server version; omitted when the platform default runs
maxmemory_policy	string	Eviction policy (see create)
persistence	object	enabled and volume size
state	string	Provisioning, Ready, Degraded, or Unknown (not yet reconciled)
ready	bool	The authoritative "you can connect now" signal
external_exposure	bool	Whether TLS external access was requested
externally_reachable	bool	Whether the external route is actually live
external_endpoint	string	rediss:// address; present only once the route is programmed
reachable_scope	string	vpc — the external endpoint is reachable from your Crusoe VPC, never the public internet
memory_limit	string	Container memory limit
maxmemory	string	The server's dataset cap (75% of the container limit)
ready_replicas, replicas	int	Always single-node in the alpha
connection.host, connection.port	string, int	In-cluster address (redis://, plaintext, port 6379)
connection.username	string	Always default
connection.credential_secret	string	The Kubernetes Secret holding the password (ms-<name>-credential)
conditions	array	CredentialReady, ServiceReady, WorkloadReady, Ready, plus ExternallyReachable for exposed instances
created_at	string	RFC 3339 timestamp

List instances

GET /v1/projects/{projectID}/memorystores

Auth: member.

Name	Type	Required	Default	Notes
page_size	int	no	50	1–200
page_token	string	no	—	Opaque cursor

Response 200:

{"memorystores": [], "next_page_token": "..."}

Get an instance

GET /v1/projects/{projectID}/memorystores/{name}

Auth: member. Returns 200 with the instance object; 404 otherwise.

Get live statistics

GET /v1/projects/{projectID}/memorystores/{name}/stats

Auth: member. Reads a live INFO snapshot directly from the running server. The internal read budget is 3 seconds.

Response 200:

{
  "memorystore": "projects/<short>/memorystores/sessions",
  "name": "sessions",
  "used_memory_bytes": 1048576, "used_memory_human": "1.00M",
  "maxmemory_bytes": 805306368, "maxmemory_policy": "noeviction",
  "used_memory_pct": 0.13,
  "connected_clients": 1, "blocked_clients": 0,
  "keyspace_hits": 10, "keyspace_misses": 2, "hit_rate": 0.8333,
  "evicted_keys": 0, "expired_keys": 0,
  "total_commands_processed": 42, "instantaneous_ops_per_sec": 0,
  "uptime_in_seconds": 3600, "db_keys": 12,
  "note": "These are live values read directly from the running instance right now - a point-in-time snapshot. ..."
}

Field behavior worth knowing:

• Fields absent from the server's INFO output are omitted, never zeroed.
• hit_rate is a 0–1 ratio and appears only after at least one lookup has happened.
• used_memory_pct appears only when maxmemory is greater than 0.
• This is a point-in-time snapshot. There is no time-series history.

Errors:

Status	Message
409	this memorystore is not provisioned yet; wait for it to report Ready, then read its stats
503	could not reach this memorystore to read its stats; it may be scaled down, still provisioning, restarting, or mid-credential-rotation - try again once it reports Ready: ...
502	could not read the memorystore stats: ...

Create an instance

POST /v1/projects/{projectID}/memorystores

Auth: member. Returns 201. Body limit: 16 KiB.

Request:

{
  "name": "sessions",
  "size_class": "medium",
  "version": "",
  "maxmemory_policy": "allkeys-lru",
  "external_exposure": false,
  "persistence": {"enabled": true, "size": "2Gi", "storage_class_name": ""}
}

Field	Type	Required	Default	Rules
name	string	yes	—	1–40 chars, ^[a-z]([-a-z0-9]*[a-z0-9])?$ (lowercase letters, digits, hyphens, starting with a letter). Becomes the DNS name; immutable
size_class	string	no	small	small, medium, or large
version	string	no	platform default	Server version; blank means the platform default
maxmemory_policy	string	no	noeviction	One of noeviction, allkeys-lru, allkeys-lfu, allkeys-random, volatile-lru, volatile-lfu, volatile-random, volatile-ttl
external_exposure	bool	no	false	Opt-in TLS rediss:// endpoint, VPC-reachable only
persistence.enabled	bool	no	true	To turn persistence off you must send {"persistence":{"enabled":false}} explicitly — omitting the field means on
persistence.size	string	no	2Gi	Volume size; fixed after create
persistence.storage_class_name	string	no	—	Kubernetes storage class override

Size classes (fixed; maxmemory is 75% of the container limit, by design, for OOM protection):

Class	Container memory	Usable (maxmemory)
small	256Mi	192Mi
medium	1Gi	768Mi
large	4Gi	3Gi

Response 201 — the only response that contains the password, besides rotate:

{
  "memorystore": {"name": "sessions", "state": "Provisioning", "ready": false},
  "credential": {
    "username": "default",
    "password": "<24-random-bytes base64url>",
    "uri": "redis://:<password>@ms-sessions.cai-p-<short>.svc.cluster.local:6379",
    "external_uri": "rediss://:<password>@sessions-<short>.<domain>:443"
  },
  "note": "This is the only response that contains the password. Later reads return the name of the Secret holding it, never the value. The instance is still provisioning; poll until state is Ready before connecting."
}

Copy it now

The password is shown once here and once per rotation. Every later read returns the name of the Kubernetes Secret that holds it (ms-<name>-credential), never the value.

external_uri appears only when external exposure is on and the endpoint is already published — rare at create time; it typically appears on a later rotate.

Errors:

Status	Message
400	invalid JSON body: ...
400	name must be 1-40 characters, lower-case letters, digits and hyphens, starting with a letter
400	sizeClass must be one of small, medium, large
400	maxmemoryPolicy must be one of ...
409	a memorystore named <name> already exists in this project
502	could not create the memorystore: ...

Delete an instance

DELETE /v1/projects/{projectID}/memorystores/{name}

Auth: admin. Returns 202:

{"name": "sessions", "resource_path": "projects/<short>/memorystores/sessions",
 "state": "Deleting",
 "note": "The instance and its data volume are being reclaimed. This is not reversible."}

Errors: 404 as usual; 502 could not delete the memorystore: ....

Rotate the credential

POST /v1/projects/{projectID}/memorystores/{name}/rotate-credential

Auth: admin. Empty body. Returns 200 with the same credential shape as create, and this note:

Rotation is not instantaneous. The server reads its password once, at start, so the new one takes effect when the instance restarts - which the platform triggers now. Until that rollout completes, the previous password is still the one in force.

Errors:

Status	Message
409	this memorystore has no credential yet; wait for it to finish provisioning
502	could not rotate the credential: ...

Health checks

GET /healthz

No auth. Readiness. Returns 200 or 503:

{"status": "ok", "checks": {
  "kubernetes": {"status": "ok"},
  "memorystore_crd": {"status": "ok"},
  "database": {"status": "ok"},
  "session_key": {"status": "ok"}}}

Failing checks report unreachable, missing, or unconfigured, plus a verbatim detail field.

GET /livez

No auth. Liveness. Always returns 200 {"status":"ok"} while the process is up.

Server limits

Limit	Value
Create body	16 KiB
Request timeout	30 s (REQUEST_TIMEOUT_SECONDS)
Stats read budget	3 s
Instance name	1–40 chars, starts with a letter
List page_size	default 50, max 200
Password appearances	create and rotate responses only

See platform limits for cross-service limits.

Status codes

Code	When
400	Invalid JSON; bad name, size class, or policy; malformed project id (projectID must be a UUID)
401	No or invalid credential
403	Member calling an admin route
404	Not found, or not yours — identical on purpose
409	Name taken; not provisioned yet (rotate, stats)
429	Per-principal rate limit
502	Upstream (Kubernetes API) failures
503	Health degraded; session signing key not loaded (this service cannot verify credentials: the shared session key is not loaded); stats unreachable

Related pages

• MemoryStore overview — what the service is and when to use it
• Connect from workloads — using the credential Secret instead of copying passwords
• CLI: data and messaging commands — the read-only platformctl memorystore commands
• API overview — shared conventions across all platform APIs