Devin Introduces Personal Memory and Daily Dreaming Process

Devin has introduced Memory and Dreaming, features designed to help the AI assistant retain useful learnings about user preferences and project context across sessions. Memory allows Devin to proactively capture corrections, preferences, and lessons learned during work, storing them as short notes in a personal Memory Drive. This drive is a persistent Git repository of markdown files, organized by repository, project, or topic, with a MEMORY.md file serving as an index. At the start of a session, Devin loads MEMORY.md and can search for relevant notes using its standard code navigation tools, rather than loading the entire archive into its prompt. Each session operates on its own Git checkout, with changes committed and merged to the persistent drive; conflicting edits are surfaced for resolution to ensure data integrity.

Dreaming is a daily background process where Devin reviews past conversations and existing memories to improve its index for future sessions. During this process, Devin consolidates overlapping notes, removes transient or stale details, and identifies useful lessons that were not captured during the original work. This ensures the memory index remains clean and relevant. The article distinguishes Memory from Skills: Skills capture repeatable workflows (procedures), while Memory captures contextual learnings (preferences and decisions). The two work together, with Skills providing the procedure and Memory providing the context for applying it. Cognition has also open-sourced the standard used to build Devin’s memory system, inviting developers to integrate it into their own agents.