Cloudflare D1 Memory

D1MemoryAdapter stores conversations in Cloudflare D1 for Workers deployments.

Installation

npm

npm install @voltagent/cloudflare-d1

yarn

yarn add @voltagent/cloudflare-d1

pnpm

pnpm add @voltagent/cloudflare-d1

Cloudflare D1 binding

Add a D1 binding to your wrangler.toml:

[[d1_databases]]
binding = "DB"
database_name = "voltagent"
database_id = "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx"

Configuration (Workers)

import { openai } from "@ai-sdk/openai";
import { Agent, Memory, VoltAgent } from "@voltagent/core";
import { D1MemoryAdapter } from "@voltagent/cloudflare-d1";
import { serverlessHono } from "@voltagent/serverless-hono";
import type { D1Database } from "@cloudflare/workers-types";

import { weatherTool } from "./tools";

type Env = {
  DB: D1Database;
  OPENAI_API_KEY: string;
};

const createWorker = (env: Env) => {
  const memory = new Memory({
    storage: new D1MemoryAdapter({
      binding: env.DB,
      tablePrefix: "voltagent_memory",
    }),
  });

  const agent = new Agent({
    name: "Assistant",
    model: openai("gpt-4o-mini"),
    tools: [weatherTool],
    memory,
  });

  const voltAgent = new VoltAgent({
    agents: { agent },
    serverless: serverlessHono(),
  });

  return voltAgent.serverless().toCloudflareWorker();
};

let cached: ReturnType<typeof createWorker> | undefined;

export default {
  fetch: (request: Request, env: Env, ctx: ExecutionContext) => {
    if (!cached) {
      cached = createWorker(env);
    }

    return cached.fetch(request, env, ctx);
  },
};

Configuration Options

Option	Type	Description
binding	D1Database	Worker D1 binding (required)
tablePrefix	string	Table name prefix (default: voltagent_memory)
maxRetries	number	Retry count for busy/locked DB operations (default: 3)
retryDelayMs	number	Initial backoff delay in ms (default: 100)
debug	boolean	Enable debug logging (default: false)
logger	Logger	Optional logger for structured logging

Features

Automatic Schema Creation

Tables are created automatically on first use:

• ${tablePrefix}_users
• ${tablePrefix}_conversations
• ${tablePrefix}_messages
• ${tablePrefix}_workflow_states
• ${tablePrefix}_steps

Conversation Storage

• Messages stored per userId and conversationId
• All StorageAdapter methods supported
• No automatic message pruning - all messages are preserved

Working Memory

Supports both conversation and user-scoped working memory:

import { z } from "zod";

const memory = new Memory({
  storage: new D1MemoryAdapter({ binding: env.DB }),
  workingMemory: {
    enabled: true,
    scope: "conversation", // or "user"
    schema: z.object({
      preferences: z.array(z.string()).optional(),
    }),
  },
});

See Working Memory for configuration details.

Semantic Search (Optional)

D1 provides storage only. To enable semantic search, pair it with an embedding + vector adapter (for example, AiSdkEmbeddingAdapter and InMemoryVectorAdapter).

See Semantic Search for usage.

Learn More

• Working Memory - Maintain compact context
• Semantic Search - Vector search configuration