@agentskit/memory β Functions
API functions for @agentskit/memory.
#Function: adaptIoredis()
adaptIoredis(
io):RedisLike
Defined in: packages/memory/src/kv-store-redis.ts:84
Bridges an ioredis-style client to the RedisLike options-object shape.
#Parameters
#io
Client with positional Redis command options.
#del
(key) => Promise<unknown>
#get
(key) => Promise<string | null>
#keys
(pattern) => Promise<string[]>
#set
(key, value, mode?, ttl?) => Promise<unknown>
#Returns
A client using the memory package's Redis contract.
#Function: chroma()
chroma(
config):VectorMemory
Defined in: packages/memory/src/vector/chroma.ts:41
Creates a vector memory backed by Chroma's HTTP API.
#Parameters
#config
Chroma URL, collection, credentials, and search defaults.
#Returns
VectorMemory
A vector memory backed by the configured collection.
#Function: createEncryptedMemory()
createEncryptedMemory(
options):Promise<ChatMemory>
Defined in: packages/memory/src/encrypted.ts:81
Wraps chat memory with AES-GCM encryption using caller-owned key material.
#Parameters
#options
Backing memory, key, and optional Web Crypto settings.
#Returns
Promise<ChatMemory>
A chat memory that encrypts saved content and decrypts loaded content.
#Throws
When the key is invalid.
#Throws
When Web Crypto is unavailable.
#Example
const key = crypto.getRandomValues(new Uint8Array(32))
const memory = await createEncryptedMemory({ backing, key })#Function: createFileStore()
createFileStore(
config):AgentskitMemoryStore
Defined in: packages/memory/src/kv-store-basic.ts:86
Creates a JSON file key-value store with atomic file replacement.
#Parameters
#config
File path and optional retention settings.
#Returns
A persistent key-value store backed by the configured file.
#Function: createHierarchicalMemory()
createHierarchicalMemory(
options):HierarchicalMemory
Defined in: packages/memory/src/hierarchical.ts:76
MemGPT-style tiered memory. Three tiers:
- working: always-loaded hot window (bounded by
workingLimit). - recall: mid-term searchable layer (usually a vector store).
- archival: cold store that always holds the full conversation.
On every save, new messages are appended to archival, messages
that overflow the working window are indexed into recall, and the
working tier is trimmed to workingLimit.
On every load, the hub returns working + up to recallTopK
messages surfaced by the recall tier, spliced chronologically.
#Parameters
#options
Working and archival memories plus optional recall settings.
#Returns
A chat memory that keeps an archival history and bounded working window.
#Function: createInMemoryGraph()
createInMemoryGraph():
GraphMemory
Defined in: packages/memory/src/graph.ts:57
In-memory graph β fine for tests, single-process demos, and as reference for what a backing store needs to implement.
#Returns
#Function: createInMemoryPersonalization()
createInMemoryPersonalization():
PersonalizationStore
Defined in: packages/memory/src/personalization.ts:28
In-memory personalization store β tests, single-process demos. Bring your own for production (Postgres, Redis, DynamoDB).
#Returns
#Function: createInMemoryStore()
createInMemoryStore(
config):AgentskitMemoryStore
Defined in: packages/memory/src/kv-store-basic.ts:61
Creates an in-memory key-value store with optional TTL and size limits.
#Parameters
#config
Backend and retention settings.
#Returns
A key-value store backed by a process-local map.
#Throws
When retention limits are invalid.
#Function: createKvMemoryFromConfig()
createKvMemoryFromConfig(
options):AgentskitMemoryStore
Defined in: packages/memory/src/kv-store-factory.ts:73
Creates a KV store for the selected backend using supplied dependencies.
#Parameters
#options
Backend config and optional database, cache, vector, and embedding adapters.
#Returns
A store implementing asynchronous get and set.
#Throws
When a selected external backend lacks a required adapter.
#Example
const store = createKvMemoryFromConfig({ config: { backend: 'in-memory' } })
await store.set('job:42', { status: 'ready' })#Function: createKvMemoryFromConfigAuto()
createKvMemoryFromConfigAuto(
config):Promise<AgentskitMemoryStore>
Defined in: packages/memory/src/kv-store-factory.ts:135
Creates a KV store and lazy-loads the optional SQLite or Redis driver.
#Parameters
#config
Backend discriminator and settings.
#Returns
Promise<AgentskitMemoryStore>
A store implementing asynchronous get and set.
#Throws
When an optional driver is missing or vector adapters are not injected.
#Function: createLocalStorageStore()
createLocalStorageStore(
options):AgentskitMemoryStore
Defined in: packages/memory/src/kv-store-basic.ts:145
Creates a local-storage store, falling back to a JSON file when unavailable.
#Parameters
#options
Storage configuration and optional storage/file adapters.
#Returns
A key-value store backed by Web Storage or the configured file.
#Throws
When retention limits are invalid.
#Function: createRedisStore()
createRedisStore(
options):AgentskitMemoryStore
Defined in: packages/memory/src/kv-store-redis.ts:23
Creates a Redis key-value store using the supplied client.
#Parameters
#options
Redis configuration and client.
#Returns
A key-value store backed by the configured Redis prefix.
#Throws
When retention limits are invalid.
#Throws
When Redis commands fail.
#Function: createSqliteStore()
createSqliteStore(
options):AgentskitMemoryStore
Defined in: packages/memory/src/kv-store-sqlite.ts:22
Creates a SQLite key-value store using the supplied database opener.
#Parameters
#options
SQLite configuration and opener.
#Returns
A key-value store backed by the configured database.
#Function: createVectorStore()
createVectorStore(
options):AgentskitMemoryStore&object
Defined in: packages/memory/src/kv-store-vector.ts:26
Creates a key-value store that embeds keys and stores values in a vector store.
#Parameters
#options
Vector config, store, and embedder.
#Returns
AgentskitMemoryStore & object
A key-value store with a similarity-based recall method.
#Throws
When the embedder returns no vector.
#Function: createWebStorageMemory()
createWebStorageMemory(
options):ChatMemory
Defined in: packages/memory/src/web-storage.ts:89
Creates a validated, bounded ChatMemory over an injected browser Web Storage backend. The storage getter is evaluated lazily so browser globals can remain SSR-safe.
#Parameters
#options
Storage key, getter, retention limits, and optional migration.
#Returns
ChatMemory
A chat memory that validates and bounds records in Web Storage.
#Throws
When the key or a configured limit is invalid.
#Throws
When a saved message record is invalid or exceeds its byte limit.
#Example
const memory = createWebStorageMemory({
key: 'agentskit-chat',
getStorage: () => globalThis.localStorage,
})#Function: fileChatMemory()
fileChatMemory(
path):ChatMemory
Defined in: packages/memory/src/file-chat.ts:24
ChatMemory backed by a JSON file on disk. Node-only.
Implements the Memory contract (ADR 0003):
- load() returns a snapshot (CM1)
- save() is replace-all, not append (CM2)
- empty state returns [] (CM5)
- clear() is optional but provided here
#Parameters
#path
string
#Returns
ChatMemory
#Function: fileVectorMemory()
fileVectorMemory(
config):VectorMemory
Defined in: packages/memory/src/file-vector.ts:116
Creates a persistent vector memory backed by a local Vectra index.
#Parameters
#config
Index path and optional vector-store adapter.
#Returns
VectorMemory
A vector memory that stores and searches local embeddings.
#Throws
When the optional vectra dependency is unavailable.
#Example
const memory = fileVectorMemory({ path: './vectors' })
await memory.store([{ id: 'guide', content: 'Guide text', embedding }])#Function: forgetSubject()
forgetSubject(
memories,subjectId):Promise<ForgetSubjectResult>
Defined in: packages/memory/src/forget.ts:85
Walk every memory passed in and run forgetSubject(subjectId) on
any that implement it. Missing capabilities are reported so callers
cannot mistake a partial deletion for a complete one.
#Parameters
#memories
unknown[]
Memory instances to inspect for deletion capability.
#subjectId
string
Subject identifier to delete.
#Returns
Promise<ForgetSubjectResult>
Per-backend reports and an incomplete flag for skipped or failed backends.
#Function: isMemoryBackendSupported()
isMemoryBackendSupported(
backend):boolean
Defined in: packages/memory/src/kv-store-factory.ts:50
Reports whether a key-value backend is currently supported.
#Parameters
#backend
"in-memory" | "file" | "sqlite" | "redis" | "vector" | "localstorage"
Backend discriminator to check.
#Returns
boolean
true when the factory can construct that backend.
#Function: makeForgettable()
makeForgettable<
M>(memory,options):M&ForgettableMemory
Defined in: packages/memory/src/forget.ts:125
Helper for backends that key records by metadata.subjectId. Wraps
any delete(ids)-style API into a ForgettableMemory.
#Type Parameters
#M
M extends object
#Parameters
#memory
M
Object to extend with the deletion capability.
#options
Backend label and functions for listing and deleting matching ids.
#backend
string
#deleteIds
(ids) => Promise<void>
#listIds
(subjectId) => Promise<string[]>
#Returns
The same object with ForgettableMemory methods and metadata.
#Function: matchesFilter()
matchesFilter(
metadata,filter):boolean
Defined in: packages/memory/src/vector/filter.ts:37
Evaluate a VectorFilter against a metadata record. Used by in-memory /
file-backed vector stores. Backends with native filter languages (pgvector,
Pinecone, Qdrant, etc.) translate the filter to their own form instead.
#Parameters
#metadata
Record<string, unknown> | undefined
#filter
VectorFilter | undefined
#Returns
boolean
#Function: milvusVectorStore()
milvusVectorStore(
config):VectorMemory
Defined in: packages/memory/src/vector/milvus.ts:36
Creates a vector memory backed by the Milvus HTTP API.
#Parameters
#config
Endpoint, collection, credentials, and search defaults.
#Returns
VectorMemory
A vector memory backed by the configured collection.
#Function: mongoAtlasVectorStore()
mongoAtlasVectorStore(
config):VectorMemory
Defined in: packages/memory/src/vector/mongo-atlas.ts:37
Creates a vector memory backed by a MongoDB Atlas collection.
#Parameters
#config
Collection adapter, vector field, and search settings.
#Returns
VectorMemory
A vector memory backed by the configured collection.
#Function: pgvector()
pgvector(
config):VectorMemory
Defined in: packages/memory/src/vector/pgvector.ts:37
Creates a vector memory backed by a PostgreSQL pgvector table.
#Parameters
#config
SQL runner and optional table and result limit.
#Returns
VectorMemory
A vector memory that upserts and searches the configured table.
#Function: pinecone()
pinecone(
config):VectorMemory
Defined in: packages/memory/src/vector/pinecone.ts:30
Creates a vector memory backed by the Pinecone vector API.
#Parameters
#config
Index endpoint, API key, and optional namespace and result limit.
#Returns
VectorMemory
A vector memory backed by the configured Pinecone index.
#Function: qdrant()
qdrant(
config):VectorMemory
Defined in: packages/memory/src/vector/qdrant.ts:54
Creates a vector memory backed by a Qdrant collection.
#Parameters
#config
Qdrant endpoint, collection, and optional credentials and result limit.
#Returns
VectorMemory
A vector memory backed by the configured collection.
#Function: redisChatMemory()
redisChatMemory(
config):ChatMemory
Defined in: packages/memory/src/redis-chat.ts:32
Creates a chat memory that stores conversation snapshots in Redis.
#Parameters
#config
Redis URL or client and optional namespace settings.
#Returns
ChatMemory
A chat memory backed by Redis.
#Throws
When the optional Redis dependency is missing or cannot connect.
#Function: redisVectorMemory()
redisVectorMemory(
config):VectorMemory
Defined in: packages/memory/src/redis-vector.ts:25
Creates a vector memory backed by Redis vector search commands.
#Parameters
#config
Redis URL or client and optional index settings.
#Returns
VectorMemory
A vector memory backed by the configured Redis index.
#Throws
When the optional Redis dependency is missing or cannot connect.
#Function: renderProfileContext()
renderProfileContext(
profile):string
Defined in: packages/memory/src/personalization.ts:64
Render a profile into a system-prompt fragment the runtime can prepend. Kept intentionally short β full profile dumps bloat context and leak unnecessary detail to the model.
#Parameters
#profile
PersonalizationProfile | null
#Returns
string
#Function: sqliteChatMemory()
sqliteChatMemory(
config):ChatMemory
Defined in: packages/memory/src/sqlite.ts:56
Creates a chat memory that stores conversation snapshots in SQLite.
#Parameters
#config
Database path and optional opener.
#Returns
ChatMemory
A chat memory backed by the configured database.
#Throws
When the optional SQLite dependency is missing or cannot open the database.
#Function: supabaseVectorStore()
supabaseVectorStore(
config):VectorMemory
Defined in: packages/memory/src/vector/supabase.ts:84
Creates a vector memory backed by a Supabase table and RPC search function.
#Parameters
#config
Supabase client, table, and optional search settings.
#Returns
VectorMemory
A vector memory backed by the configured Supabase schema.
#Function: tryDefaultRedisClient()
tryDefaultRedisClient(
url):Promise<RedisLike|undefined>
Defined in: packages/memory/src/kv-store-redis.ts:102
Loads and connects node-redis, returning undefined when it is unavailable.
#Parameters
#url
string
Redis connection URL.
#Returns
Promise<RedisLike | undefined>
A connected client, or undefined when the optional package is absent.
#Throws
When the client cannot connect.
#Function: tryDefaultSqliteOpener()
tryDefaultSqliteOpener():
Promise<SqliteOpener|undefined>
Defined in: packages/memory/src/kv-store-sqlite.ts:84
Lazy-import better-sqlite3 and return an opener, or undefined when the
optional peer dep is absent (caller surfaces AK_MEMORY_PEER_MISSING).
#Returns
Promise<SqliteOpener | undefined>
An opener, or undefined when better-sqlite3 is unavailable.
#Function: tursoChatMemory()
tursoChatMemory(
config):ChatMemory
Defined in: packages/memory/src/turso.ts:70
libSQL / Turso-backed chat memory. Mirrors the sqliteChatMemory shape so
code paths can swap between local SQLite and replicated libSQL by changing
one import.
@libsql/client is an optional peer dependency loaded lazily.
#Parameters
#config
#Returns
ChatMemory
#Function: upstashVector()
upstashVector(
config):VectorMemory
Defined in: packages/memory/src/vector/upstash.ts:30
Upstash Vector β HTTP-only serverless vector DB. The REST surface is tiny enough to implement directly without pulling the SDK.
#Parameters
#config
#Returns
VectorMemory
#Function: weaviateVectorStore()
weaviateVectorStore(
config):VectorMemory
Defined in: packages/memory/src/vector/weaviate.ts:36
Creates a vector memory backed by a Weaviate class.
#Parameters
#config
Weaviate endpoint, class, and optional credentials and result limit.
#Returns
VectorMemory
A vector memory backed by the configured class.
#Function: wrapChatMemoryWithRedaction()
wrapChatMemoryWithRedaction(
inner,options):ChatMemory
Defined in: packages/memory/src/redaction.ts:89
Wraps chat memory to redact or tokenize message content before saves.
#Parameters
#inner
ChatMemory
Chat memory implementation to wrap.
#options
Redaction rules and optional tokenization settings.
#Returns
ChatMemory
A chat memory wrapper that redacts saved message content.
#Throws
When tokenization lacks a vault or allowed roles.
#Function: wrapVectorMemoryWithRedaction()
wrapVectorMemoryWithRedaction(
inner,options):VectorMemory
Defined in: packages/memory/src/redaction.ts:126
Wrap any VectorMemory so each document's content is redacted (or
tokenized) before store(). search() and delete() pass through.
Note: embeddings pass through verbatim. Customers who embed plaintext PII separately (e.g. via a hosted embedding provider) must redact the input to their embedder, not just to this wrapper.
#Parameters
#inner
VectorMemory
#options
#Returns
VectorMemory