REA turns application observations into places for agents to work
REA 6.8.0 turns application captures into local environments where AI agents can read derived state and make changes supported by recorded behavior. The release also allows comparisons with fresh captures. Its maintainers describe changed defaults for environment serving and function analysis, alongside interface changes that require users to revisit existing integrations. Command-line users retain complete output, while MCP clients receive compact input schemas by default.
Artificial Intelligence··Midday
Captured behavior defines the environment
REA 6.8.0, released on 25 October, introduces a route from application observations to local environments for AI agents. The list of tools and their behavior is reconstructed in manifests using network captures alongside evidence files. Initial state can come from what a browser scenario observes, or from a prompt. The resulting environments can be exported as AgentEnv bundles and compared with fresh application captures to examine their agreement with observed behavior.[1]
Reads follow state and writes follow evidence
The environment server now defaults to derived state. Search, filtering, sorting and pagination read from that state, while changes through write tools are restricted to effects evidenced in application recordings. Users can still explicitly choose replay without state. Older browser-storage captures can still be read. Comparing them requires recapturing both sides in the changed representation; until then, the comparison stays unknown. These are maintainer-described behavior changes in a single release document.[1]
Analysis output changes its default
Function analysis returns a retained summary by default, with complete inline detail available on request. The command-line interface keeps its complete-output default and requires an evidence-output path when a summary is selected. MCP input schemas also become compact by default, with full guidance available through configuration. The release flags breaking interface changes, updates Node.js requirements and leaves runtime upgrades to the user rather than performing them automatically.[1]
