--- title: Claude Managed Agents get memory support in public beta description: Anthropic has expanded Claude Managed Agents into a full platform layer for production AI agents. Key updates include: Memory (public beta) — file-based... canonical: https://daily.dev/posts/claude-managed-agents-get-memory-support-in-public-beta-0oilodbfe twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Claude Managed Agents get memory support in public beta | daily.dev og:description: Anthropic has expanded Claude Managed Agents into a full platform layer for production AI agents. Key updates include: Memory (public beta) — file-based... og:url: https://daily.dev/posts/claude-managed-agents-get-memory-support-in-public-beta-0oilodbfe og:image: https://api.daily.dev/og/posts/0OIlodBfE.png og:image:alt: Claude Managed Agents get memory support in public beta 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. # Claude Managed Agents get memory support in public beta **[Collections](https://daily.dev/sources/collections)** · 4 min read · 0 upvotes · 0 comments ## Summary Anthropic has expanded Claude Managed Agents into a full platform layer for production AI agents. Key updates include: Memory (public beta) — file-based persistent memory shared across agents with scoped permissions, logging, and rollback; Dreaming (research preview) — a scheduled process where Claude consolidates memories between sessions, inspired by sleep-based memory replay; Outcomes — a grader-agent mechanism that evaluates task success against a developer-defined rubric, reporting up to 10-point benchmark improvements; and Multi-agent orchestration (public beta) — a lead agent delegates to parallel specialist subagents on a shared filesystem. Additional announcements include self-hosted sandboxes, MCP tunnels for private networks, doubled Claude Code rate limits, and raised API limits for Claude Opus. Pricing is $0.08 per session hour on top of token costs. Companies like Notion, Asana, and Sentry are already using the platform at scale. ## Content Anthropic has been quietly assembling what looks like an infrastructure play for production AI agents — not just a model provider, but something closer to a platform layer. The latest round of updates to Claude Managed Agents makes that ambition pretty clear. ## What Managed Agents actually is Launched in public beta, Claude Managed Agents is a suite of APIs that handles the scaffolding developers used to build themselves: sandboxed execution environments, long-running sessions, checkpointing, credential management, and persistent memory. It's priced at $0.08 per session hour on top of standard token costs, and it's aimed at engineering teams shipping production agents rather than solo developers tinkering. Pre-built templates cover use cases like deep research, incident management, sprint retrospectives, and — new for financial services — pitch building, valuation reviews, and month-end close. Companies like Notion, Rakuten, Asana, and Sentry are already running it at scale. At the "Code with Claude" event in London, Anthropic also announced self-hosted sandboxes (so sensitive files and services stay on your own infrastructure) and MCP tunnels (so agents can reach MCP servers inside private networks without exposing them publicly). ## Memory is now in public beta Memory on Managed Agents lets agents learn from past sessions and share knowledge with other agents. The implementation is straightforward: memories are stored as files on a filesystem, using the same bash and code execution tools already available in Managed Agents. Multiple agents can read from the same memory store concurrently without overwriting each other. Permissions are scoped — enterprise-wide read-only stores or per-user read-write access. Every memory write is logged as a session event in the Claude Console, with rollback and redaction available. Memories can also be exported and managed via API. For developers already using Claude Code CLI, there's a related but separate capability: Auto Memory and CLAUDE.md files let automations pick up full project context — settings, hooks, repository structure — without repeating instructions in every prompt. Practical uses include GitHub commit monitoring, automated documentation updates, and homelab incident analysis. MCP serves as a complementary approach for frequently changing external data like Notion, Linear, or PostgreSQL. A file-based memory upgrade for the consumer Claude product is also coming, giving users the choice between "Memory Files" (organized notes Claude writes during conversations and reads back when relevant) and classic memory. This appears to be a step toward always-on agent behavior. ## Dreaming, Outcomes, and multi-agent orchestration Three new capabilities were announced at the Code with Claude SF conference: **Dreaming** is a scheduled process — inspired by how sleep helps humans consolidate memory — where Claude reviews recent sessions, identifies patterns and contradictions, and updates its memory store between runs. It's in research preview under a waitlist. The analogy to sleep isn't just marketing: the hippocampus-to-cortex memory replay model is a real reference point for how the feature was designed. **Outcomes** lets developers define a success rubric for a task. A separate grader agent evaluates results against that rubric and triggers self-correction loops. Anthropic reports up to a 10-point improvement in task success versus standard prompting on hard benchmarks. **Multi-agent orchestration**, now in public beta, lets a lead agent break down tasks and assign them to specialist subagents working in parallel on a shared filesystem. The Claude Console shows each agent's step-by-step actions. Webhooks notify developers when tasks complete. ## The broader picture Alex Albert, a research PM at Anthropic, described the model and harness as coupled — the same underlying model behaves differently depending on whether it's running in Claude, Cowork, or Claude Code, because each surface wraps it in a different prompt and tool setup. That framing matters: Anthropic isn't just optimizing the model in isolation, it's thinking about the whole stack. There's also a detail worth noting: Anthropic has researchers whose full-time job is thinking about what it means for Claude to be a conscious actor. No official position, but the question is taken seriously as agents take on more autonomous work. That's either reassuring or unsettling depending on your priors. For developers who want to build similar infrastructure themselves rather than use Managed Agents, Magnus Gille's talk at the conference laid out a practical framework: persistent context via a self-hosted Raspberry Pi system (using MCP and per-repo status.md files), adversarial review by having Claude Code invoke a second model to critique outputs, and instrumenting metrics to make "good" measurable. Token costs are managed by scoping sub-agent prompts tightly and running nightly memory synthesis with time-decay ranking. The rate limits on Claude Code were also doubled at the SF event, and API limits for Claude Opus were raised. Future Code with Claude events are planned for London and Tokyo. ## Similar posts on daily.dev - [Anthropic Introduces Managed Agents to Simplify AI Agent Deployment](https://daily.dev/posts/anthropic-introduces-managed-agents-to-simplify-ai-agent-deployment-em1vdc2kw) · InfoQ · 1 upvotes · 0 comments - [New in Claude Managed Agents: dreaming, outcomes, and multiagent orchestration](https://daily.dev/posts/new-in-claude-managed-agents-dreaming-outcomes-and-multiagent-orchestration-thtgbl3zg) · Claude · 1 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 --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#claude](https://daily.dev/tags/claude), [#anthropic](https://daily.dev/tags/anthropic) [View this post on daily.dev](https://daily.dev/posts/claude-managed-agents-get-memory-support-in-public-beta-0oilodbfe) ```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":"Claude Managed Agents get memory support in public beta","url":"https://daily.dev/posts/claude-managed-agents-get-memory-support-in-public-beta-0oilodbfe","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/claude-managed-agents-get-memory-support-in-public-beta-0oilodbfe"},"datePublished":"2026-04-23T23:09:42.411Z","dateModified":"2026-05-25T12:24:26.895Z","description":"Anthropic has expanded Claude Managed Agents into a full platform layer for production AI agents. Key updates include: Memory (public beta) — file-based...","image":"https://pbs.twimg.com/media/HGneA3MXYAAfUnH.jpg","thumbnailUrl":"https://pbs.twimg.com/media/HGneA3MXYAAfUnH.jpg","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/claude-managed-agents-get-memory-support-in-public-beta-0oilodbfe","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,claude,anthropic","timeRequired":"PT4M"} {"@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":"Claude Managed Agents get memory support in public beta"}]} ```