Skip to main content
Attribution filtering checks the code Devin generates against a corpus of publicly available code before it is written to your codebase. When a match against public code is detected, the write is blocked and Devin rewrites the implementation instead. This gives enterprises additional assurance that generated code does not reproduce public sources with incompatible licenses.

​
How it works

When attribution filtering is enabled for your organization, every edit Devin makes through its file-editing tools in a cloud session — from creating new files to modifying existing ones — is checked before it executes:
  1. The code snippets Devin is about to write are extracted from the pending edit. Trivial snippets that are too short to meaningfully match public code are skipped.
  2. Any secrets available to the session are scrubbed from the snippets before they leave the session.
  3. The snippets are checked against Cognition’s public-code attribution service.
  4. If a match against public code is found, the edit is rejected and Devin is instructed to rewrite the implementation in its own words or use a permissively licensed source with proper attribution.
Attribution filtering fails closed: if the attribution service cannot be reached, or the snippets cannot be safely scrubbed of secrets, the edit is blocked rather than allowed through unchecked. Attribution filtering is also available for the Devin Local agent (Devin CLI) as a team-wide enterprise setting — see Devin CLI team settings.

​
Enabling attribution filtering

Attribution filtering is available on Enterprise plans and is not enabled by default. There is currently no self-serve toggle — to enable it for your organization, reach out to support@cognition.ai or contact your Cognition account team.
Attribution filtering adds a check before code is written, which can add small amounts of latency to file edits. Sessions remain fully functional; Devin simply rewrites any code that matches public sources.