Replit

Replit started as a browser IDE and now centers on Replit Agent. You describe an app; the agent sets up the environment, installs dependencies, writes and runs the code, and tests it in a preview, all on Replit’s cloud. Larger requests become tasks on a board (Drafts, Active, Ready, Done). Background tasks run in isolated copies of the project and change nothing until you apply them: Core runs 1 at a time, Pro up to 10, Enterprise up to 64. Free Mode picks a model for you within a plan allowance; Power and Max Modes use stronger models and are billed per checkpoint, priced by how much work the request took.

Every app gets two PostgreSQL databases. The agent can create and change tables in the development database but cannot touch production; schema changes are applied to production when you publish. Checkpoints snapshot the files, the conversation, the agent’s memory and the development database, so you can roll everything back, optionally including the data. Rollbacks don’t restore the production database; that takes a point-in-time restore.

Publishing takes a snapshot of the app and runs it as Autoscale (scales to zero, billed while serving requests), Static, Reserved VM (always on, fixed monthly cost) or Scheduled (cron). Published apps run on Google Cloud in the United States, each customer in its own single-tenant GCP project; Enterprise customers can ask for the European Union. Core costs $20 a month and includes $20 of credits for Agent, publishing and databases; Pro costs $100 and includes $100.