Skip to main content

Cassandra KV

本示例演示了如何通过 CassandraKVStore 的 BaseStore 集成来设置聊天历史记录存储。请注意，还有一个 CassandraChatMessageHistory 集成，使用起来可能更简单，适合聊天历史记录存储。如果您需要一个支持前缀键的通用键值存储，则可以使用 CassandraKVStore。

安装配置​

npm install @langchain/community @langchain/core cassandra-driver

根据您的数据库提供商，连接数据库的具体方法会有所不同。我们将创建一个 configConnection 文档，它将用于存储配置。

Apache Cassandra®​

在 Apache Cassandra® 5.0 及以上版本中支持存储附加索引（由 yieldKeys 使用）。您可以使用标准连接文档，例如：

const configConnection = {
  contactPoints: ['h1', 'h2'],
  localDataCenter: 'datacenter1',
  credentials: {
    username: <...> as string,
    password: <...> as string,
  },
};

Astra DB​

Astra DB 是一个云原生的 Cassandra 即服务（Cassandra-as-a-Service）平台。

  1. 创建一个 Astra DB 账户。
  2. 创建一个启用向量的数据库。
  3. 为您的数据库创建一个token。
const configConnection = {
  serviceProviderArgs: {
    astra: {
      token: <...> as string,
      endpoint: <...> as string,
    },
  },
};

除了使用 endpoint: 属性外，您也可以使用 datacenterID: 属性，并可选地使用 regionName: 属性。

使用方法​

import { CassandraKVStore } from "@langchain/community/storage/cassandra";
import { AIMessage, HumanMessage } from "@langchain/core/messages";

// This document is the Cassandra driver connection document; the example is to AstraDB but
// any valid Cassandra connection can be used.
const configConnection = {
  serviceProviderArgs: {
    astra: {
      token: "YOUR_TOKEN_OR_LOAD_FROM_ENV" as string,
      endpoint: "YOUR_ENDPOINT_OR_LOAD_FROM_ENV" as string,
    },
  },
};

const store = new CassandraKVStore({
  ...configConnection,
  keyspace: "test", // keyspace must exist
  table: "test_kv", // table will be created if it does not exist
  keyDelimiter: ":", // optional, default is "/"
});

// Define our encoder/decoder for converting between strings and Uint8Arrays
const encoder = new TextEncoder();
const decoder = new TextDecoder();

/**
 * Here you would define your LLM and chat chain, call
 * the LLM and eventually get a list of messages.
 * For this example, we'll assume we already have a list.
 */
const messages = Array.from({ length: 5 }).map((_, index) => {
  if (index % 2 === 0) {
    return new AIMessage("ai stuff...");
  }
  return new HumanMessage("human stuff...");
});

// Set your messages in the store
// The key will be prefixed with `message:id:` and end
// with the index.
await store.mset(
  messages.map((message, index) => [
    `message:id:${index}`,
    encoder.encode(JSON.stringify(message)),
  ])
);

// Now you can get your messages from the store
const retrievedMessages = await store.mget(["message:id:0", "message:id:1"]);
// Make sure to decode the values
console.log(retrievedMessages.map((v) => decoder.decode(v)));

/**
[
  '{"id":["langchain","AIMessage"],"kwargs":{"content":"ai stuff..."}}',
  '{"id":["langchain","HumanMessage"],"kwargs":{"content":"human stuff..."}}'
]
 */

// Or, if you want to get back all the keys you can call
// the `yieldKeys` method.
// Optionally, you can pass a key prefix to only get back
// keys which match that prefix.
const yieldedKeys = [];
for await (const key of store.yieldKeys("message:id:")) {
  yieldedKeys.push(key);
}

// The keys are not encoded, so no decoding is necessary
console.log(yieldedKeys);
/**
[
  'message:id:2',
  'message:id:1',
  'message:id:3',
  'message:id:0',
  'message:id:4'
]
 */

// Finally, let's delete the keys from the store
await store.mdelete(yieldedKeys);

API Reference:

相关内容​

Was this page helpful?


You can also leave detailed feedback on GitHub.