Introduction

db0 is a TypeScript SDK that gives AI agents structured memory, recoverable state, and token-budgeted context assembly. It runs local-first on SQLite with no API keys required, and scales to PostgreSQL when you need cross-device sync.

Why db0

Every team building agents ends up rebuilding the same infrastructure: a memory store, some way to recover from bad states, a system for stuffing relevant context into the token window. db0 packages all of that into a single harness API.

• Memory — scoped key-value store with vector search, fact superseding, and relationship edges
• State — checkpoint, branch, and restore execution state
• Context — ingest documents, pack memories into token budgets, preserve facts before compaction
• Log — structured, append-only event logging
• Sub-agents — spawn child harnesses with shared user memory and isolated task memory

How it works

Everything goes through a harness — a scoped handle to the storage layer:

import { db0 } from "@db0-ai/core"
import { createSqliteBackend } from "@db0-ai/backends-sqlite"

const backend = await createSqliteBackend({ dbPath: "./agent.db" })
const harness = db0.harness({
  agentId: "my-agent",
  sessionId: "session-1",
  userId: "user-123",
  backend,
})

The harness scopes all operations to a specific agent, session, and user. Memory visibility follows a 4-level hierarchy: task < session < user < agent.

Packages

Core

Integrations

Tools

Next

• Getting started — install and write your first memory
• Vercel AI SDK — add memory to the AI SDK in one line
• LangChain — replace deprecated BufferMemory
• Tutorial — build a chat agent with persistent memory