Memory Configuration

Configure agent memory — auto-capture, retrieval, namespaces, and embedding models.

AgentMemoryConfig

interface AgentMemoryConfig {
  store: MemoryStore;
  config: MemoryConfig;
}

interface MemoryConfig {
  autoCapture?: boolean;        // Auto-save key facts from runs
  retrieveBeforeStep?: boolean; // Retrieve relevant memories before each LLM call
  maxRetrieved?: number;        // Max memories to retrieve (default: 5)
  namespace?: string;           // Isolate memories by namespace
  embedding?: EmbeddingConfig;  // Embedding model configuration
}

Auto-Capture

When autoCapture: true, AgentForge automatically extracts and stores key facts from completed runs. The agent learns from every interaction.

memory: {
  store: memoryStore,
  config: {
    autoCapture: true,  // Learn automatically
  },
},

Retrieve Before Step

When retrieveBeforeStep: true, relevant memories are retrieved and injected into the LLM context before each step:

memory: {
  store: memoryStore,
  config: {
    retrieveBeforeStep: true,
    maxRetrieved: 10,  // Top 10 most relevant memories
  },
},

Namespaces

Isolate memories by namespace (e.g., per-customer):

memory: {
  store: memoryStore,
  config: {
    namespace: 'customer_123',
  },
},

Embedding Configuration

interface EmbeddingConfig {
  provider?: string;    // Embedding provider
  model?: string;       // Embedding model
  dimensions?: number;  // Vector dimensions
}

Next Steps

In-Memory Store — development memory
PgVector Store — production memory
Embeddings — embedding models