@prisma/studio-core

@prisma/studio-core

@prisma/studio-core is the embeddable Prisma Studio package.

It provides the same core experience as Prisma Studio: a visual way to explore schema, browse table data, edit rows, filter/sort/paginate records, inspect relation data, and run SQL queries with an operation log.

This package is published to npm and consumed by Prisma surfaces such as Console and CLI integrations.

Embedding Studio

Import the UI entrypoint, include the packaged CSS once, and pass a configured adapter into Studio:

import { Studio } from "@prisma/studio-core/ui";
import "@prisma/studio-core/ui/index.css";
import { createStudioBFFClient } from "@prisma/studio-core/data/bff";
import { createPostgresAdapter } from "@prisma/studio-core/data/postgres-core";

const adapter = createPostgresAdapter({
  executor: createStudioBFFClient({
    url: "/api/query",
  }),
});

export function EmbeddedStudio() {
  return (
    <Studio
      adapter={adapter}
      aiFilter={async (request) => {
        const response = await fetch("/api/ai-filter", {
          body: JSON.stringify({ prompt: request }),
          headers: { "content-type": "application/json" },
          method: "POST",
        });

        const payload = (await response.json()) as {
          error?: unknown;
          text?: unknown;
        };

        if (!response.ok) {
          throw new Error(
            typeof payload.error === "string"
              ? payload.error
              : `AI filter request failed (${response.status} ${response.statusText})`,
          );
        }

        if (typeof payload.text !== "string") {
          throw new Error("AI filter response did not include text.");
        }

        return payload.text;
      }}
    />
  );
}

adapter is required. aiFilter is optional and is the new LLM completion hook: it takes the user's natural-language request and returns JSON text describing one or more filters. When aiFilter is omitted, Studio renders only the manual filter controls and hides the inline AI prompt entirely. Studio no longer renders a built-in fullscreen header button. If your host needs fullscreen behavior, render that control at the host container level, as the local demo does. Studio handles prompt construction, type-aware validation, correction retries, and conversion into the normal URL-backed filter pills. The hook can return regular column filters or a SQL fallback when the request cannot be expressed with the predefined operators.

Integration Checklist

Studio is an embeddable React surface, not a standalone app shell. A production integration should:

• render <Studio /> inside the host product's route, panel, or page
• import @prisma/studio-core/ui/index.css exactly once
• create one database adapter per connection using createPostgresAdapter, createMySQLAdapter, or createSQLiteAdapter
• back that adapter with an authenticated executor, typically createStudioBFFClient({ url: "/api/query" })
• expose a JSON BFF endpoint that accepts Studio requests and executes them against the database
• pass any tenant or auth context through customHeaders and/or customPayload
• optionally provide aiFilter for natural-language filtering
• own surrounding product chrome such as routing, auth, tenancy, and fullscreen controls

The simplest supported shape is: host React app -> <Studio /> -> adapter -> createStudioBFFClient(...) -> host BFF route -> database executor.

BFF Contract

Studio's packaged adapters speak one JSON-over-HTTP contract. The host application is expected to implement a POST endpoint, usually /api/query, that accepts StudioBFFRequest payloads and returns JSON results with serialized errors.

Yes: we did change this contract to support staged multi-cell saves. The current contract includes procedure: "transaction", which is used to commit multiple staged row updates in one database transaction when the backend supports it.

Transport Rules

• method: POST
• content type: application/json
• request authentication: whatever the host app requires; createStudioBFFClient forwards customHeaders as HTTP headers
• request body context: customPayload is forwarded in the JSON body unchanged
• error encoding: return serialized errors, not thrown JS objects

Shared Types

type Query = {
  sql: string;
  parameters: readonly unknown[];
  transformations?: Partial<Record<string, "json-parse">>;
};

type SerializedError = {
  name: string;
  message: string;
  errors?: SerializedError[];
};

type SqlLintDiagnostic = {
  code?: string;
  from: number;
  message: string;
  severity: "error" | "warning" | "info" | "hint";
  source?: string;
  to: number;
};

Request Shapes

type StudioBFFRequest =
  | {
      procedure: "query";
      query: Query;
      customPayload?: Record<string, unknown>;
    }
  | {
      procedure: "sequence";
      sequence: readonly [Query, Query];
      customPayload?: Record<string, unknown>;
    }
  | {
      procedure: "transaction";
      queries: readonly Query[];
      customPayload?: Record<string, unknown>;
    }
  | {
      procedure: "sql-lint";
      sql: string;
      schemaVersion?: string;
      customPayload?: Record<string, unknown>;
    };

Response Shapes

type QueryResponse = [SerializedError, undefined?] | [null, unknown[]];

type SequenceStepResponse = [SerializedError] | [null, unknown[]];

type SequenceResponse =
  | [SequenceStepResponse]
  | [[null, unknown[]], SequenceStepResponse];

type TransactionResponse = [SerializedError, undefined?] | [null, unknown[][]];

type SqlLintResponse =
  | [SerializedError, undefined?]
  | [
      null,
      {
        diagnostics: SqlLintDiagnostic[];
        schemaVersion?: string;
      },
    ];

Procedure Semantics

• query: execute one SQL statement. This is required for every Studio adapter.
• sequence: execute exactly two queries in order. This is used by MySQL write flows that update first and refetch second.
• transaction: execute an ordered list of queries inside one database transaction. This is the contract addition that enables atomic staged multi-row saves from the table editor.
• sql-lint: return parse/plan diagnostics for the SQL editor and SQL-backed filter pills.

For sequence, the second query should only run if the first one succeeds. For transaction, the response result array must stay in the same order as body.queries.

sql-lint is optional because adapters can fall back to adapter-local EXPLAIN strategies. transaction is strongly recommended because it gives staged multi-row saves atomic behavior; without it, adapters may fall back to sequential writes.

Example BFF Handler

The demo server in this repo is the reference implementation. A host route can mirror it closely:

import { serializeError, type StudioBFFRequest } from "@prisma/studio-core/data/bff";

export async function handleStudioBff(request: Request): Promise<Response> {
  if (request.method !== "POST") {
    return new Response("Method Not Allowed", {
      headers: { Allow: "POST,OPTIONS" },
      status: 405,
    });
  }

  const payload = (await request.json()) as StudioBFFRequest;

  if (payload.procedure === "query") {
    const [error, result] = await executor.execute(payload.query);
    return Response.json([error ? serializeError(error) : null, result]);
  }

  if (payload.procedure === "sequence") {
    const [firstQuery, secondQuery] = payload.sequence;
    const [firstError, firstResult] = await executor.execute(firstQuery);

    if (firstError) {
      return Response.json([[serializeError(firstError)]]);
    }

    const [secondError, secondResult] = await executor.execute(secondQuery);

    if (secondError) {
      return Response.json([
        [null, firstResult],
        [serializeError(secondError)],
      ]);
    }

    return Response.json([
      [null, firstResult],
      [null, secondResult],
    ]);
  }

  if (payload.procedure === "transaction") {
    if (typeof executor.executeTransaction !== "function") {
      return new Response("Transaction execution is not supported", {
        status: 501,
      });
    }

    const [error, result] = await executor.executeTransaction(payload.queries);
    return Response.json([error ? serializeError(error) : null, result]);
  }

  if (payload.procedure === "sql-lint") {
    if (typeof executor.lintSql !== "function") {
      return new Response("SQL lint is not supported", { status: 501 });
    }

    const [error, result] = await executor.lintSql({
      schemaVersion: payload.schemaVersion,
      sql: payload.sql,
    });

    return Response.json([error ? serializeError(error) : null, result]);
  }

  return new Response("Invalid procedure", { status: 400 });
}

Integration Notes

• If your host app changes auth or tenant context at runtime, recreate the BFF client or adapter so new customHeaders and customPayload are used for later requests.
• If you are embedding MySQL Studio, keep sequence support enabled because the adapter depends on ordered write-plus-refetch flows.
• If you want fully atomic staged table saves, implement transaction on the BFF and forward it to a real database transaction on the server.
• If you omit aiFilter, Studio still supports the full manual filtering UI.

Telemetry

This package includes anonymized telemetry to help us improve Prisma Studio. Set CHECKPOINT_DISABLE=1 to opt out of usage-data collection, following Prisma's documented CLI telemetry opt-out contract. Learn more in our Privacy Policy and the Prisma CLI telemetry docs.

Run Studio Locally

Requirements:

• Node.js ^20.19 || ^22.12 || >=24.0
• pnpm 8
• bun

Install dependencies and start the demo:

pnpm install
pnpm demo:ppg

Then open http://localhost:4310.

To enable the demo's AI filtering flow, copy .env.example to .env and set ANTHROPIC_API_KEY. The demo reads that key server-side and calls Anthropic Haiku 4.5 directly over HTTP for the aiFilter hook. Set STUDIO_DEMO_AI_FILTERING_ENABLED=false to hide the AI filter UI without removing the key. .env and .env.local are gitignored.

The demo:

• starts Prisma Postgres dev (ppg-dev) programmatically via @prisma/dev
• uses direct TCP for query execution
• seeds sample relational data on startup
• auto-rebuilds and reloads the UI when source files change

The demo database is intentionally ephemeral: it is pre-seeded when the demo starts and reset when the demo process stops.

Useful Commands

• pnpm demo:ppg - run local Studio demo with seeded Prisma Postgres dev
• pnpm typecheck - run TypeScript checks
• pnpm lint - run ESLint (--fix)
• pnpm test - run default vitest suite
• pnpm test:checkpoint - run checkpoint tests
• pnpm test:data - run data-layer tests
• pnpm test:demo - run demo/server tests
• pnpm test:ui - run UI tests
• pnpm test:e2e - run e2e tests
• pnpm demo:ppg:build - bundle the demo server with bun build
• pnpm demo:ppg:bundle - build and run the bundled demo server
• pnpm build - build distributable package with tsup
• pnpm check:exports - validate package export map/types

When bundling the demo with bun build, we use --packages external so @prisma/dev can resolve its PGlite runtime assets (WASM/data/extensions) directly from node_modules at runtime.

Development Workflow

For day-to-day development, run the demo locally and verify both terminal logs and browser behavior as you make changes.

Recommended flow:

• Run pnpm demo:ppg.
• Keep the demo process attached so rebuilds and runtime logs stay visible.
• Validate UI behavior at http://localhost:4310, using Playwright when you want an automated browser check.

Because the demo is pre-seeded and resets between runs, update seed data whenever needed to reproduce richer scenarios.

Seed data lives in demo/ppg-dev/seed-database.ts (seedDatabase).