How AI Agents Use Parallel for Testing

The Problem

Agents need to test code that uses databases, caches, and message queues. Current options:

• Mock everything (unreliable, doesn't catch real bugs)
• Use real services (slow, expensive, side effects)
• Skip testing (dangerous)

The Solution

Parallel gives agents a lightweight gym to practice against real protocols.

Agent Experience

Option 0: Remote sandbox — just an API key + a sidecar (no app changes)

The leanest agent path: add the Parlel bridge sidecar to your stack and point your app at it. The bridge provisions a sandbox from a pk_ key and exposes every dependency on a plain hostname — Postgres/Redis over a WebSocket tunnel (the sandbox's HTTPS preview proxy can't carry raw TCP), Jira/Stripe/etc. via a token-injecting reverse-proxy. Your application code is unchanged: it connects to parlel-bridge:<port> exactly as it would to a real service in production. No SSH tunnel, no daytona CLI, no Parlel code in your app.

# docker-compose.test.yml
include:
  - path: parlel-bridge.yml          # image: parlel/bridge
services:
  app:
    build: .
    env_file: test.env               # DATABASE_URL=postgres://parlel:parlel@parlel-bridge:5432/parlel
    depends_on:
      parlel-bridge: { condition: service_healthy }

PARLEL_API_KEY=pk_... docker compose -f docker-compose.test.yml up

Your app's unmodified drivers connect straight through:

import psycopg, redis  # unmodified, real wire protocol
db = psycopg.connect("postgres://parlel:parlel@parlel-bridge:5432/parlel")
cache = redis.Redis(host="parlel-bridge", port=6379)

A sandbox with no services specified defaults to Postgres + Redis. See the Flask CRUD example for the full sidecar flow, and Agent Access for the API + bridge details.

Option 1: Drive setup over MCP

Every sandbox exposes an MCP endpoint. Agents seed data and inspect state with the parlel_execute tool — no driver needed:

{
  "name": "parlel_execute",
  "arguments": {
    "service": "postgres",
    "command": "CREATE TABLE users (id serial, email text); SELECT * FROM users;"
  }
}

Then exercise your application code against the same sandbox through the bridge sidecar (Option 0). See Agent Access for the full API.

What Agents Get

Example: testing a user service

Your test suite is whatever you already use — it just runs against the bridge hostnames. Nothing Parlel-specific in the code:

# conftest.py / fixtures point at the bridge; the app reads its normal env vars
import os, psycopg, redis
from user_service import UserService

def test_create_and_cache_user():
    db = psycopg.connect(os.environ["DATABASE_URL"])     # parlel-bridge:5432
    cache = redis.from_url(os.environ["REDIS_URL"])        # parlel-bridge:6379
    svc = UserService(db, cache)

    user = svc.create_user("alice@test.com", "Alice")
    assert user.id
    assert "Alice" in cache.get(f"user:{user.id}")

Run it inside the compose stack (docker compose -f docker-compose.test.yml run app pytest) or against a sandbox you provisioned over the API.

Architecture

┌──────────────────────────────────────────────┐
│              Your app / tests                  │   (unchanged, plain env vars)
├──────────────────────────────────────────────┤
│   DATABASE_URL → parlel-bridge:5432            │
│   REDIS_URL    → parlel-bridge:6379            │
├──────────────────────────────────────────────┤
│         parlel/bridge  (Docker sidecar)        │
│   TCP-over-WS  │  HTTP reverse-proxy + token   │
├──────────────────────────────────────────────┤
│      Parlel sandbox (hosted, ephemeral)        │
│   Postgres │ Redis │ Kafka │ Stripe │ Jira …   │
└──────────────────────────────────────────────┘

Why This Works

1. Real protocols - Not mocks. Real Redis RESP, Postgres/MySQL wire protocol, Kafka binary protocol (real kafkajs produces and consumes), and AMQP 0-9-1 (real amqplib publishes and consumes).
2. Ephemeral - State disappears when tests end.
3. Fast - Starts in <1 second.
4. Small - ~220KB vs ~880MB for real services.
5. Composable - Use only what you need.
6. Remote with just a key - In a sandbox, the parlel/bridge sidecar lets unmodified drivers connect over an API key — no SSH, no CLI, no app changes.