A SQL database per agent,
per app, per PR.

Spin up an isolated SQLite database in milliseconds. Pay only for the requests, rows, and storage you use.

Start free Agent quickstart

Typical latency

—

Use cases

A database for every unit of work

One call creates a database — made on demand, isolated by default, and you only pay when it's used.

Per agent

Every run gets its own database. db.asTool() hands it to the model as a tool — mistakes stay contained.

Per PR

Every pull request gets a branch — your tables, no production rows — that cleans itself up after merge.

Per app

Every microservice and prototype gets its own database instead of a corner of a shared Postgres.

Coding agents

Drop it into your coding agent

Your agent provisions its own database the first time it needs one. Sign in once in the browser — no API key to copy.

terminal

# Claude Code
claude mcp add --transport http persql https://mcp.persql.com/mcp

# Codex
codex mcp add persql --url https://mcp.persql.com/mcp

Claude Code, Codex, Cursor, OpenCode — any MCP client. Install guide

SDK

Hand your agent a database

TypeScript and Python. One key turns any database into a tool your agent can call. Hand sub-agents their own branch, with a role and an expiry.

agent.ts

import { PerSQL } from "@persql/sdk";

const db = new PerSQL({ token: process.env.PERSQL_TOKEN! })
  .database("acme/agent-7af");

// One isolated DB, handed to the model as a tool.
const tools = [db.asTool().anthropic]; // or .openai

handoff.ts

// Lease a scoped, expiring DB to a sub-agent.
const { token } = await db.branches.pin("pr-42", {
  role: "write",
  ttlSec: 900,
});

const sub = await PerSQL.fromHandoff(token);
const { namespaceSlug, databaseSlug } = sub.handedOff;
await sub.database(namespaceSlug, databaseSlug)
  .query("INSERT INTO leads (email) VALUES (?)", ["a@b.co"]);

State for your framework — a LangGraph checkpointer and OpenAI Agents sessions ship as packages.

GitHub

A preview database on every pull request

One step in your workflow. Each PR gets a branch — your tables, no production rows — that cleans itself up.

.github/workflows/preview.yml

steps:
  - uses: persql/preview-db-action@v1
    with:
      token: ${{ secrets.PERSQL_TOKEN }}
      database: acme/orders
      branch: pr-${{ github.event.pull_request.number }}
      ttl-seconds: 86400 # one day, then gone

Read the recipe

Pricing

Only what you use

Prepaid and usage-based. Usage draws down a balance you load, and spending stops at zero. No seats. No per-database fees. New accounts start with $5 in credit.

Per pull request

~$0.01

A preview database on every PR. Copy your tables, run migrations and tests, delete it on merge — about 500 requests and 55,000 rows.

Per agent session

~$0.02

An isolated database per run. 1,000 queries reading 100,000 rows and writing 10,000.

Production app

~$21/mo

10M requests, 1B rows read, 5M rows written, 2 GB stored.

Estimated from the meter rates below.

Requests	$0.50	per million
Rows read	$0.005	per million
Rows written	$2.00	per million
Storage	$0.40	per GB / month
AI tokens	$1.00	per million

All included on every account. Features aren't tiers.

Create your first database

Start with $5 in credit. Provision from the console, the CLI, or an agent.

Start free Read the docs

npm i @persql/sdk
pip install persql
brew install persql/tap/persql
docker run persql/cli

A SQL database per agent, per app, per PR.