REA Environments is a workflow in the open source REA toolkit for teams that train and evaluate agents with reinforcement learning. It takes a recorded session of a site or app and turns it into an environment Scale's AgentEnv framework can deploy, run, and grade. The release follows Scale Labs open sourcing AgentEnv on October 5, 2026, the framework every RL environment Scale builds already runs on. REA, at rea.tools, connects a coding agent to local inspection of native binaries, JavaScript and Electron apps, and browser activity, and it keeps the evidence behind each finding. A saved HAR file or a scripted browser scenario is projected into an environment manifest. From that manifest, REA serves an MCP server whose tools replay recorded responses, or exports a bundle AgentEnv can build and run, including a Dockerfile, a Python verifier, and a smoke task. Tool calls are answered from the capture. The server does not contact the live application, and it does not reproduce the application's authorization rules or state changes. The launch points to 50 replay environments REA built from recorded browser sessions of 43 self-hosted apps and 7 public APIs, exposed as MCP tools on AgentEnv. The set includes Gitea, Grafana, Sentry, Ghost, Mattermost, and Immich, generally run self-hosted on synthetic data, with retrieval tasks attached. Notes on each environment list the app version, license, tool count, and retrieval scores for gpt-oss-120b and GLM-5.3-Flash. Scale engineer Edgar Arakelyan published the collection and was credited in the launch thread.