--- title: Cognition's Devin adds "Dreaming" for memory cleanup,... description: Cognition has added a feature called Dreaming to Devin, its coding agent, which runs an offline cleanup pass over the agent's memory graph each night, removing... canonical: https://daily.dev/posts/cognition-s-devin-adds-dreaming-for-memory-cleanup-and-an-open-memory-standard-ebuhqtsfn twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Cognition's Devin adds "Dreaming" for memory cleanup, and an open memory standard | daily.dev og:description: Cognition has added a feature called Dreaming to Devin, its coding agent, which runs an offline cleanup pass over the agent's memory graph each night, removing... og:url: https://daily.dev/posts/cognition-s-devin-adds-dreaming-for-memory-cleanup-and-an-open-memory-standard-ebuhqtsfn og:image: https://api.daily.dev/og/posts/EbUHqTSFn.png og:image:alt: Cognition's Devin adds "Dreaming" for memory cleanup, and an open memory standard og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Cognition's Devin adds "Dreaming" for memory cleanup, and an open memory standard **[Collections](https://daily.dev/sources/collections)** · 2 min read · 1 upvotes · 0 comments ## Summary Cognition has added a feature called Dreaming to Devin, its coding agent, which runs an offline cleanup pass over the agent's memory graph each night, removing stale records and surfacing latent information. Cognition is also launching an open-source standard called Agent Memory Repo for agent memory. LangChain's Harrison Chase praised the offline cleanup approach as addressing a real gap (stale retrieval), noted memory is hard to get right in company settings, and welcomed the standard being open rather than vendor-locked. An unresolved question is how inferred memories get validated before an agent acts on them, since a bad inference or wrongful deletion could cause more harm than a messy memory store. ## Content Cognition has introduced Dreaming, a memory feature for Devin. Across sessions, Devin builds a memory graph of how a user likes to work. At night, it goes back over that memory, removes stale records and looks for latent information it hadn't surfaced before. Cognition is also creating an open-source standard called Agent Memory Repo. ## Why the offline pass matters @hwchase17 pointed to the part of this that interests me most: memory needs an offline cleanup loop, not just better retrieval. Without one, stale records pile up and the agent ends up retrieving things that are no longer true. Better search over a messy store doesn't fix that. He also said memory is hard to get working well, especially in a company setting, and that he is making changes to how memory works in managed deepagents. ## The open question Dreaming lets the agent infer things on its own. That raises the obvious problem: how are those inferred memories validated before the agent acts on them? A cleanup loop that deletes the wrong record, or promotes a bad inference, could do more harm than an untidy store. Cognition's announcement doesn't answer this. ## Open by default @hwchase17 said he is glad the standard is open, adding that memory should be open. I agree. If an agent's memory of how you work is locked inside one vendor's product, you can't take it with you. ## Questions this post answers ### What is the Dreaming feature in Cognition's Devin agent? Dreaming is an offline memory maintenance process for Devin that runs at night, going back over the memory graph Devin builds from user sessions. It removes stale records that are no longer true and surfaces latent information that hadn't been explicitly noted before, addressing the problem of agents retrieving outdated memories rather than just improving search over a messy store. _Developers evaluating AI coding agents can follow how memory handling evolves across tools on daily.dev._ ### What is the Agent Memory Repo standard Cognition announced? Agent Memory Repo is an open-source standard for agent memory that Cognition is creating alongside the Dreaming feature for Devin. The intent is to keep an agent's memory of how a user works portable rather than locked inside a single vendor's product, so that memory data isn't stranded if someone switches tools. _Teams weighing vendor lock-in for AI agent tooling can track open standards like this one on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 3 discussions and 172 comments across x (as of 2026-10-06). **TL;DR:** Reaction is largely enthusiastic about git-versioned, diffable agent memory and the open standard, but a recurring thread of skepticism centers on whether inferred memories get validated before an agent trusts them, and whether 'dreaming' might silently delete or hallucinate structure that was never really there. **Sentiment:** 45% positive · 40% mixed · 15% skeptical **The case for** - Storing memory as git-versioned files that can be diffed is widely seen as a strong, 'boring but useful' primitive. - Nightly pruning of stale records is viewed as solving a real problem since memory that only grows gets noisier over time. - People like that the standard is open rather than vendor-locked and plan to try or adapt it themselves. - Several found early use promising, with concrete examples of duplicate/stale notes being merged successfully. **The pushback** - Many worry the same model that writes an inferred memory will also be the one 'validating' it, effectively rubber-stamping its own guesses. - Concern that deletions during dreaming should be reviewable (like a PR/commit) rather than silent, with some wanting tombstones or deletion reasons. - Doubt about scope: a one-off instruction or personal habit could wrongly get promoted into a permanent or team-wide preference. - Questions about how the system handles conflicting memories across team members or across changing project configs (e.g., SQLite to Postgres). - Some feel deleting outdated entries loses useful history, preferring to keep both old and new states rather than erasing. **By community** - x (mixed): Replies mix genuine enthusiasm for git-based, prunable agent memory with persistent pushback on how inferred memories get validated and whether cleanup should ever silently delete rather than propose/flag changes. **Hottest debate:** Whether an agent can be trusted to validate and prune its own inferred memories, or whether promotion/deletion requires a separate check (human or second model) to avoid the agent simply rubber-stamping its own guesses. **Open questions** - How exactly is memory effectiveness measured, and how is staleness determined (access patterns, semantic clustering, agent judgment)? - How are personal, repo, and team/org-level memories separated and who approves promotion from personal to shared? - Can users see, restore, or pin specific memories that dreaming chose to prune? - How does the system handle two sessions or team members disagreeing on a deletion or preference? - Does cleanup propagate to conclusions already built on a memory that later gets corrected or deleted? **Highlights** > @cognition Memory in Git is the right primitive. Agent memory you can't diff is memory you can't trust. The risky part is the "dreaming": when Devin prunes a "stale" record overnight, I want that as a reviewable commit, not a silent rewrite. Memory changes deserve PRs too. > — [hashif360 on x](https://x.com/hashif360/status/2107184245462671510) > @KolinHuang @cognition Pinning is the right instinct, and it points at the real design: the cleanup shouldn't delete, only propose. A newer note replaces an older one by naming it, and the old one stays on the record with its words. Then nothing needs pinning, because nothing can vanish without a > — [Magnemo\_AI on x](https://x.com/Magnemo_AI/status/2107273373625856236) > @cognition The edge case I'd test is a one-off instruction becoming a global preference. Memory needs scope: this repo, this task, or all projects. With two repos using conflicting conventions, does retrieval select the right preference before the agent plans? > — [scott\_stouffer on x · 1 points](https://x.com/scott_stouffer/status/2107191713664004198) > @cognition The removal half is the one the standard should pin down, because memory also has a read-side cost. In our runtime the invocation context packs the memory block in with config, cwd, event sinks, instructions, MCP registry and hooks, one field among seven. A nightly job that only > — [echo\_vic on x](https://x.com/echo_vic/status/2107199281341812788) > @cognition The dreaming agent spots patterns across sessions and rewrites memory. That is sort of a compression step; what checks keep it from hallucinating patterns that were never really there? > — [kartmehra on x · 1 comments](https://x.com/kartmehra/status/2107229099194548381) **Source threads** - [x](https://x.com/trikcode/status/2107206305790255552) · 0 points · 0 comments - [x](https://x.com/cognition/status/2107165034463867001) · 0 points · 125 comments - [x](https://x.com/hwchase17/status/2107206402078851519) · 0 points · 47 comments ## Similar posts on daily.dev - [Anthropic gave agents the ability to dream. Then developers woke up.](https://daily.dev/posts/anthropic-gave-agents-the-ability-to-dream-then-developers-woke-up--lsyubx30x) · The New Stack · 0 upvotes · 0 comments - [Claude's leaked dreaming feature is now live, and it lets agents learn from their own mistakes](https://daily.dev/posts/claude-s-leaked-dreaming-feature-is-now-live-and-it-lets-agents-learn-from-their-own-mistakes-48xpxeqxf) · XDA Developers · 2 upvotes · 0 comments - [Memory Hygiene Beats Memory Volume: Why AI Agents Should Forget Things](https://daily.dev/posts/memory-hygiene-beats-memory-volume-why-ai-agents-should-forget-things-tuy8tflfr) · Medium · 0 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#langchain](https://daily.dev/tags/langchain), [#devin](https://daily.dev/tags/devin) [View this post on daily.dev](https://daily.dev/posts/cognition-s-devin-adds-dreaming-for-memory-cleanup-and-an-open-memory-standard-ebuhqtsfn) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Cognition's Devin adds \"Dreaming\" for memory cleanup, and an open memory standard","url":"https://daily.dev/posts/cognition-s-devin-adds-dreaming-for-memory-cleanup-and-an-open-memory-standard-ebuhqtsfn","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/cognition-s-devin-adds-dreaming-for-memory-cleanup-and-an-open-memory-standard-ebuhqtsfn"},"datePublished":"2026-10-05T20:29:12.457Z","dateModified":"2026-10-06T07:00:22.893Z","description":"Cognition has added a feature called Dreaming to Devin, its coding agent, which runs an offline cleanup pass over the agent's memory graph each night, removing...","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/cognition-s-devin-adds-dreaming-for-memory-cleanup-and-an-open-memory-standard-ebuhqtsfn","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,langchain,devin","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"Cognition's Devin adds \"Dreaming\" for memory cleanup, and an open memory standard"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/cognition-s-devin-adds-dreaming-for-memory-cleanup-and-an-open-memory-standard-ebuhqtsfn#faq","mainEntity":[{"@type":"Question","name":"What is the Dreaming feature in Cognition's Devin agent?","acceptedAnswer":{"@type":"Answer","text":"Dreaming is an offline memory maintenance process for Devin that runs at night, going back over the memory graph Devin builds from user sessions. It removes stale records that are no longer true and surfaces latent information that hadn't been explicitly noted before, addressing the problem of agents retrieving outdated memories rather than just improving search over a messy store. Developers evaluating AI coding agents can follow how memory handling evolves across tools on daily.dev."}},{"@type":"Question","name":"What is the Agent Memory Repo standard Cognition announced?","acceptedAnswer":{"@type":"Answer","text":"Agent Memory Repo is an open-source standard for agent memory that Cognition is creating alongside the Dreaming feature for Devin. The intent is to keep an agent's memory of how a user works portable rather than locked inside a single vendor's product, so that memory data isn't stranded if someone switches tools. Teams weighing vendor lock-in for AI agent tooling can track open standards like this one on daily.dev."}}]} ```