AI SDK Projects

Create SaaS-scoped SDK projects in the dashboard and avoid tenantId vs projectId confusion.

Use this page first if you are wiring memories.sh into a SaaS app.

SaaS apps using SDK endpoints (/api/sdk/v1/*) do not need MCP multitenancy. Use MCP routing only when you are serving MCP-native clients.

What is an AI SDK Project?

In the dashboard, an AI SDK Project is a tenantId mapping to an isolated memory database.

One project per customer workspace, org, environment, or app surface
Works for solo devs and teams
Created in your currently selected dashboard workspace

Setup Flow (Dashboard)

Open Dashboard → AI SDK Projects.
Generate a mem_... API key.
Create your first project:
- Attach Existing DB (recommended): map an existing Turso database.
- Provision New Project DB: create a new Turso DB from the dashboard.
Use that tenantId from your backend when calling memories.

Scope Cheat Sheet

Scope field	What it isolates	Security boundary	Required
`tenantId`	SaaS customer/project database	Yes	For multi-tenant SaaS
`userId`	End-user scope inside `tenantId`	Yes (within tenant)	Optional
`projectId`	Optional git/repo context filter	No	Optional

Important: `tenantId` vs `projectId`

These are different:

tenantId = SaaS isolation boundary (database routing)
userId = end-user scope inside tenantId
projectId = optional repo context filter (for coding assistants)

If your app is not git-centric, omit projectId.

Solo Dev Guidance

You do not need a team workspace to start.

Stay in your personal workspace.
Create AI SDK Projects there.
Add team workspace later if needed.

Server-Side Usage Pattern

Always derive scope in backend auth code, then call memories with your server key.

const client = new MemoriesClient({
  apiKey: process.env.MEMORIES_API_KEY,
  tenantId: auth.orgId, // server-derived
  userId: auth.userId,  // server-derived
})

const context = await client.context.get({
  query: "pricing and billing policy",
  projectId: undefined, // optional git context filter
})

Next: SaaS Auth Routing