Bloom vs Replit AI

These two don't compete. Bloom is a research tool for testing how AI models behave (does a model flatter users, resist shutdown, and so on), and Replit is a browser-based place to build and host apps with an AI agent. Most people land here searching for a Replit-hosted app called something like "Bloomie Days"; that app isn't on AIDiveForge, and bloomiedays.replit.app returned a 404 on 18 October 2026.

Bloom is for AI safety researchers

Bloom is an open-source Python package (MIT) that writes evaluation suites for one behavior you name, runs them against a target model and scores the transcripts, in four stages: Understanding, Ideation, Rollout and Judgment. You install it with pip, add your own model API keys, and run it from the command line. It was published under the safety-research GitHub organization, but the repo now says it's frozen at its last standalone release and won't get updates. Development moved to Meridian Labs as Petri Bloom (pip install petri-bloom). If you want Bloom today, start there.

Replit is for building apps

Replit runs in the browser: you describe an app, its Agent writes and runs the code, and you publish it on a replit.app address. The "bloomiedays.replit.app" searches are looking for one of those published apps, not for Replit itself. The pricing page today lists Core at $20 a month ($18 a month billed annually) with $20 toward model usage, Pro at $100 a month ($90 billed annually) with $100 toward models and up to 15 collaborators, and custom Enterprise pricing. It's closed source and you can't self-host it.