Vercel AI SDK

@db0-ai/ai-sdk adds persistent memory to the Vercel AI SDK via middleware. The SDK is stateless by default — db0 injects memories before each LLM call and extracts facts after, with zero extra LLM calls.

npm · GitHub

Install

npm i @db0-ai/ai-sdk

Quick start

import { createDb0 } from "@db0-ai/ai-sdk"
import { generateText } from "ai"
import { anthropic } from "@ai-sdk/anthropic"

const memory = await createDb0({ dbPath: "./agent.db" })

const result = await generateText({
  model: anthropic("claude-sonnet-4-20250514"),
  middleware: memory.middleware,
  prompt: "My name is Alex, I work on backend systems.",
})

On the next call, db0 automatically recalls that the user is Alex and works on backend systems — no code changes needed.

How it works

The middleware hooks into two points in the AI SDK lifecycle:

1. Before the LLM call (transformParams): extracts facts from the user message, searches for relevant memories, and injects them into the system prompt within a token budget.
2. After the LLM call (wrapGenerate): extracts facts from the assistant response and stores them.

Extraction is rules-based. No extra LLM calls.

Three usage modes

Middleware only (automatic)

Memory is invisible to the model. Facts are extracted and recalled automatically.

const memory = await createDb0()

const result = await generateText({
  model: anthropic("claude-sonnet-4-20250514"),
  middleware: memory.middleware,
  prompt: "Remember that I prefer TypeScript over Python.",
})

Tools only (agent-controlled)

The model decides when to read and write memories via tool calls.

const result = await generateText({
  model: anthropic("claude-sonnet-4-20250514"),
  tools: memory.tools,
  prompt: "What do you know about me?",
})

Both (hybrid)

Middleware handles automatic extraction; tools give the model explicit control when needed.

const result = await generateText({
  model: anthropic("claude-sonnet-4-20250514"),
  middleware: memory.middleware,
  tools: memory.tools,
  prompt: "Save this: I'm migrating from Express to Hono.",
})

Memory tools

When using tools mode, the model gets three tools:

Configuration

const memory = await createDb0({
  dbPath: "./agent.db",         // default: ./db0.sqlite
  agentId: "my-agent",          // default: ai-sdk
  userId: "user-123",           // default: "default"
  tokenBudget: 1500,            // tokens for memory injection
  extractOnResponse: true,      // extract facts from responses
})

For PostgreSQL, pass a backend instead of dbPath:

import { createPostgresBackend } from "@db0-ai/backends-postgres"

const backend = await createPostgresBackend({
  connectionString: "postgresql://user:pass@host:5432/db",
})

const memory = await createDb0({ backend })

Session management

Call newSession() to start a fresh conversation while keeping all memories:

memory.newSession("session-2")

Providers

Works with any Vercel AI SDK provider: Anthropic, OpenAI, Google, Mistral, Cohere, and others. The middleware wraps the model, not the provider.