Classes

AgentMemoryClient

Agent Memory Client

Manages conversation memory using Redis Agent Memory Server (AMS)

Constructor

constructor(config: AgentMemoryClientConfig)

Parameters:

Methods

healthCheck

Check if Agent Memory Server is healthy

healthCheck(): Promise<boolean>

Returns:

Promise<boolean> - Promise<boolean> - True if healthy, false otherwise

readWorkingMemory

Read working memory from AMS

readWorkingMemory(sessionId: string, namespace?: string | undefined): Promise<AmsMemory | null>

Parameters:

Returns:

Promise<AmsMemory \| null> - Promise<AmsMemory | null> - Memory object or null if not found

replaceWorkingMemory

Replace working memory in AMS

Uses PUT /v1/working-memory/{session_id}

replaceWorkingMemory(sessionId: string, contextWindowMax: number, memory: AmsMemory, namespace?: string | undefined): Promise<AmsMemory | null>

Parameters:

Returns:

Promise<AmsMemory \| null> - Promise<AmsMemory | null> - Updated memory object or null on failure

removeWorkingMemory

Remove working memory from AMS

Uses DELETE /v1/working-memory/{session_id}?namespace=xxx

removeWorkingMemory(sessionId: string, namespace?: string | undefined): Promise<boolean>

Parameters:

Returns:

Promise<boolean> - Promise<boolean> - True if removed successfully, false otherwise

addMessage

Add a message to AMS and get updated memory

This method:

1. GETs existing memory for the session
2. Appends the new message to the messages array
3. PUTs the updated memory back to AMS

AMS will automatically summarize if context window is exceeded.

addMessage(sessionId: string, namespace: string, role: AmsRole, content: string, contextWindowMax?: number | undefined): Promise<AmsMemory | null>

Parameters:

Returns:

Promise<AmsMemory \| null> - Promise<AmsMemory | null> - Updated memory object or null on failure

getMemoryPrompt

Get memory-enhanced prompt ready for LLM

Uses AMS Memory Prompt API to build messages with conversation context POST /v1/memory/prompt

getMemoryPrompt(params: MemoryPromptRequest): Promise<MemoryPromptResponse | null>

Parameters:

Returns:

Promise<MemoryPromptResponse \| null> - Promise<MemoryPromptResponse | null> - Messages array or null on failure

listSessions

List all sessions for a namespace

GET /v1/working-memory?namespace=xxx

listSessions(namespace: string, options?: ListSessionsOptions | undefined): Promise<SessionListResponse>

Parameters:

Returns:

Promise<SessionListResponse> - Promise<SessionListResponse> - List of sessions

createLongTermMemories

Create one or more long-term memories

POST /v1/long-term-memory/

createLongTermMemories(memories: CreateLongTermMemoryParams[]): Promise<LongTermMemory[]>

Parameters:

Returns:

Promise<LongTermMemory[]> - Promise<LongTermMemory[]> - Created memories

createLongTermMemory

Create a single long-term memory

Convenience wrapper around createLongTermMemories

createLongTermMemory(params: CreateLongTermMemoryParams): Promise<LongTermMemory | null>

Parameters:

Returns:

Promise<LongTermMemory \| null> - Promise<LongTermMemory | null> - Created memory or null on failure

searchLongTermMemory

Search long-term memories

POST /v1/long-term-memory/search

searchLongTermMemory(params: SearchLongTermMemoryParams): Promise<SearchLongTermMemoryResponse>

Parameters:

Returns:

Promise<SearchLongTermMemoryResponse> - Promise<SearchLongTermMemoryResponse> - Search results

forgetMemories

Forget (delete) long-term memories based on policy

POST /v1/long-term-memory/forget

forgetMemories(params: ForgetMemoriesParams): Promise<ForgetMemoriesResponse>

Parameters:

Returns:

Promise<ForgetMemoriesResponse> - Promise<ForgetMemoriesResponse> - Deletion results

deleteLongTermMemory

Delete a specific long-term memory by ID

Convenience method that uses forgetMemories with a specific memory ID filter

deleteLongTermMemory(memoryId: string, namespace: string): Promise<boolean>

Parameters:

Returns:

Promise<boolean> - Promise<boolean> - True if deleted successfully

searchAllMemory

Search across all memory types (working memory and long-term memory)

Performs parallel searches across both memory stores and combines results. For working memory, performs client-side text matching on session messages. For long-term memory, uses semantic vector search via AMS API.

searchAllMemory(params: SearchAllMemoryParams): Promise<SearchAllMemoryResponse>

Parameters:

Returns:

Promise<SearchAllMemoryResponse> - Promise<SearchAllMemoryResponse> - Combined search results

Examples:

const results = await client.searchAllMemory({
  query: 'user preferences for dark mode',
  namespace: 'org-123',
  includeLongTermMemory: true,
  includeWorkingMemory: true,
  limit: 10,
});

results.results.forEach(result => {
  console.log(`[${result.source}] ${result.text} (score: ${result.score})`);
});

buildContextPrompt

Build a formatted context prompt from memory

Useful for constructing prompts with conversation history

buildContextPrompt(memory: AmsMemory): string

Parameters:

Returns:

string - string - Formatted context prompt

getConfig

Get the current configuration

getConfig(): Readonly<Required<Omit<AgentMemoryClientConfig, "authToken">> & Pick<AgentMemoryClientConfig, "authToken">>

Returns:

Readonly<Required<Omit<AgentMemoryClientConfig, "authToken">> & Pick<AgentMemoryClientConfig, "authToken">> - AgentMemoryClientConfig - Current configuration

Examples:

const client = new AgentMemoryClient({
  baseUrl: 'http://agent-memory-server:8000',
  contextWindowMax: 10000,
});

// Check health
const isHealthy = await client.healthCheck();

// Add a message
const memory = await client.addMessage(
  'session-123',
  'default',
  AmsRole.USER,
  'Hello, how are you?'
);

// Read working memory
const currentMemory = await client.readWorkingMemory('session-123', 'default');