Anthropic Managed Agents — Memory

https://platform.claude.com/docs/en/managed-agents/overview

Mounts memory as files on a managed filesystem; Claude reads and writes via the same bash/code-execution tools used for agentic tasks. Unit of storage is a file — inspectable, editable, exportable by developers. Full audit log per session. Memory beta launched April 23, 2026, two weeks after Managed Agents itself launched April 8.

At a glance

Type
Filesystem-backed cross-session persistent memory
Tier
T3
Created
2026-04-08 (Managed Agents GA); 2026-04-23 (Memory beta added)
Latest release
not applicable — not OSS
License
not applicable — not OSS
GitHub
not applicable — no GitHub repo
Pricing
Claude API token rates + $0.08 per session-hour for Managed Agents
Funding
Anthropic $12.4B+ raised; $40B valuation

Taxonomy

storage
file
retrieval
extraction-pull
persistence
cross-session
update
agent-controlled
unit
file
governance
inspectable
conflict
none

When to use

Optimised for: filesystem-as-memory + tool-driven persistence

Anti-fit: not for non-Claude providers

Pros & cons

Pros

Filesystem metaphor aligns with Claude's existing tool-use primitives; no separate embedding pipeline; human-auditable without special tooling.

Cons

Memory granularity and eviction policy not yet documented publicly; filesystem-as-memory has known scalability limits at large knowledge volumes; beta status means API surface may change.

Claims & capabilities

Rakuten reported 97% fewer first-pass errors, 27% lower cost, 34% lower latency in early-adopter usage.

Technical surface

API surface
not applicable — research paper
Backend storage
not applicable — research paper
Deployment
Managed-only (hosted Anthropic infrastructure)
Embedding model
not applicable — research paper
Multi-tenancy
not applicable — research paper
MCP
native (first-party) — Anthropic created MCP
A2A
not supported
OpenTelemetry
OpenTelemetry traces via Anthropic Console + OTLP export for Enterprise

Similar systems

Other platform-provider memory in the catalog, ranked by inbound references.

  • AWS Bedrock AgentCore Memory T2

    AWS managed memory service for agents running on Bedrock AgentCore. Native integrations with LangChain/LangGraph and Strands. Drives long-running MCP servers + agent workloads.

  • Anthropic Claude Memory T1

    Memory tool API + consumer "Memory" feature + Auto Dream consolidation. Document-as-memory rather than extracted-fact-as-memory.

  • Cloudflare Agent Memory T2

    Managed persistent memory service for AI agents on Cloudflare's edge. Beta-launched April 2026. Sits alongside Workers AI / Vectorize as the memory tier of the Cloudflare agent stack.

  • Gemini Enterprise Agent Platform Memory Bank (rebrand of Vertex AI Memory Bank) T1

    GA at Google Cloud Next 2026 (2026-04-22). Vertex AI rebranded as Gemini Enterprise Agent Platform; Memory Bank now "Agent Platform Memory Bank" with new continuous event-streaming + automated memory generation triggered by configurable criteria (event count or idle time). Backed by Agent Runtime that supports long-running agents that maintain state for days. Note: separate "Memory Profiles" feature reported in TheNextWeb but not surfaced in official Google Cloud release notes — treated here as unconfirmed.

  • Google Gemini Memory T1

    Personal Context setting + Personal Intelligence packing. Memory import from ChatGPT/Claude (Mar 2026).

  • Grok Memory (xAI) T1

    Extracts and persists facts from past conversations; applies them to future responses. Per-memory deletion UI; user can also clear all.

Related systems

References (1)

Row last verified 2026-05-14. Catalog data is CC-BY-4.0 — see how to read this.