LangChain.js Integration

This page covers @dakera-ai/langchain, the LangChain.js package for Dakera. It is for TypeScript and Node.js developers who want LangChain chains and retrievers backed by Dakera's persistent agent memory and server-side embeddings.

The current release is @dakera-ai/langchain 0.3.0. It builds on the TypeScript SDK (@dakera-ai/dakera 0.12.1 or later, peer dependency ^0.12.1) and @langchain/core 1.1 or later.

Quick start

npm install @dakera-ai/langchain @dakera-ai/dakera @langchain/core

Requires Node.js 20 or later and a running Dakera server (v0.12.0 or v0.11.108) (see the Quickstart).

import { DakeraMemory, DakeraVectorStore } from "@dakera-ai/langchain";

const memory = new DakeraMemory({
  apiUrl: "http://localhost:3000",
  apiKey: "dk-mykey",
  agentId: "my-agent",
});

const vectorstore = new DakeraVectorStore({
  apiUrl: "http://localhost:3000",
  apiKey: "dk-mykey",
  namespace: "my-docs",
});

Exports

DakeraMemory: extends LangChain's BaseMemory; recalls relevant memories for each turn and stores the exchange
DakeraVectorStore: extends VectorStore for RAG, with embeddings computed by the Dakera server; also offers hybridSearch and fulltextSearch
DakeraSessionManager: starts and ends sessions for one agent
DakeraKnowledgeGraph: knowledge graph operations
DakeraEntityExtractor: entity extraction from text
DakeraNamespaceManager: namespace management
DakeraAgentTools: agent statistics and memory listing

Examples

Conversation chain with persistent memory

With LangChain.js v1, the legacy ConversationChain lives in @langchain/classic: npm install @langchain/classic @langchain/openai.

import { ChatOpenAI } from "@langchain/openai";
import { ConversationChain } from "@langchain/classic/chains";
import { DakeraMemory } from "@dakera-ai/langchain";

const memory = new DakeraMemory({
  apiUrl: "http://localhost:3000",
  apiKey: "dk-mykey",
  agentId: "chat-agent",
  recallK: 5,
  importance: 0.7,
});

const chain = new ConversationChain({
  llm: new ChatOpenAI({ model: "gpt-4o" }),
  memory,
});

await chain.invoke({ input: "My name is Alice." });

// Later, in another process: the memory is still on the Dakera server
const res = await chain.invoke({ input: "What's my name?" });

RAG retriever

import { DakeraVectorStore } from "@dakera-ai/langchain";

const vectorstore = new DakeraVectorStore({
  apiUrl: "http://localhost:3000",
  apiKey: "dk-mykey",
  namespace: "my-docs",
});

// Add documents (IDs are generated when you do not pass { ids })
await vectorstore.addDocuments([
  { pageContent: "Dakera computes embeddings on the server.", metadata: {} },
  { pageContent: "Memory decay lowers the importance of stale memories.", metadata: {} },
]);

// Retrieve
const retriever = vectorstore.asRetriever({ k: 4 });
const docs = await retriever.invoke("How does Dakera embed text?");

Session management

import { DakeraSessionManager } from "@dakera-ai/langchain";

const sessions = new DakeraSessionManager({
  apiUrl: "http://localhost:3000",
  apiKey: "dk-mykey",
  agentId: "my-agent",
});

const sessionId = await sessions.start({ user: "alice" }); // metadata; returns the session ID
// ... pass sessionId to memory operations ...
await sessions.end(); // ends the active session

end() accepts a summary argument but 0.3.0 does not send it to the server.

API reference

DakeraMemory options

Parameter	Type	Default	Description
`apiUrl`	string	—	Dakera server URL (required)
`apiKey`	string	—	API key
`agentId`	string	—	Agent whose memories are read and written (required)
`recallK`	number	`5`	Memories to surface per turn
`minImportance`	number	`0`	Minimum importance of recalled memories
`importance`	number	`0.7`	Importance assigned to stored memories
`memoryKey`	string	`"history"`	Key injected into the prompt
`inputKey`	string	—	Input key to read the user message from

DakeraVectorStore options

Parameter	Type	Default	Description
`apiUrl`	string	—	Dakera server URL (required)
`apiKey`	string	—	API key
`namespace`	string	—	Vector namespace to read and write (required)

Configuration

The package does not read environment variables; pass apiUrl and apiKey to each constructor. To keep them out of source code, read them yourself:

const memory = new DakeraMemory({
  apiUrl: process.env.DAKERA_API_URL ?? "http://localhost:3000",
  apiKey: process.env.DAKERA_API_KEY,
  agentId: "my-agent",
});