@anvia/memory-prisma: Getting Started

Install

pnpm add @anvia/memory-prisma @anvia/core @prisma/client

Generate the Prisma models

Run a dry run first:

npx @anvia/memory-prisma init

Write the generated model file:

npx @anvia/memory-prisma init --write

The default write path creates prisma/models/anvia-memory.prisma. If your project cannot use multi-file Prisma schemas, append explicitly:

npx @anvia/memory-prisma init --write --append-to-schema

The append path warns before writing because it modifies the existing schema.prisma.

After writing schema changes:

npx prisma validate
npx prisma migrate dev --name add_anvia_memory

Configure memory

import { AgentBuilder } from "@anvia/core";
import { createPrismaMemoryStore } from "@anvia/memory-prisma";
import { prisma } from "./db";

const memory = createPrismaMemoryStore(prisma, {
  scope: {
    metadataKeys: ["tenantId"],
  },
});

const agent = new AgentBuilder("support", model)
  .memory(memory, { savePolicy: "turn" })
  .build();

await agent
  .session("thread_123", {
    userId: "user_456",
    metadata: { tenantId: "tenant_789" },
  })
  .prompt("Where is my order?")
  .send();

scope defines the database key for one memory thread. The default key includes sessionId and userId; metadataKeys: ["tenantId"] adds metadata.tenantId, which isolates memory across tenants or workspaces.

Next step

Continue with Usage Patterns.