memU journal

Ideas, releases, and the evolution of memU

Product decisions, implementation notes, and practical thinking about personal memory across agents.

Articles marked Earlier-version content are preserved for historical context and may not describe the current memU architecture or setup.

316 articles

Cover for memU Gives the Ability to Remember Back to the Agent Itself — Across Sessions, Across Agents, Across Devices

Current memU

memU Team

Jul 17featured

memU Gives the Ability to Remember Back to the Agent Itself — Across Sessions, Across Agents, Across Devices

The harness has moved into the model. memU's July release deletes the pipelines, prompts, and LLM retrieval chains: the agent decides what to remember; memU just stores, embeds, and retrieves — in 474 lines, with zero LLM calls.

Read article

Cover for Anthropic Agent Skills Turn Markdown into Reusable Claude Playbooks — But Packaged Skills Are Not Company-Wide Memory

Earlier-version content

MemU Team

Apr 01featured

Anthropic Agent Skills Turn Markdown into Reusable Claude Playbooks — But Packaged Skills Are Not Company-Wide Memory

Claude Agent Skills package runbooks and guardrails as reusable files. Static skills still need MemU for memory of real outcomes across teams.

Read article

Cover for Claude Code Evaluation Harnesses Score Agent PRs and Diffs — But Harness Metrics Without Memory Retest the Same Edge Cases

Earlier-version content

MemU Team

Apr 01featured

Claude Code Evaluation Harnesses Score Agent PRs and Diffs — But Harness Metrics Without Memory Retest the Same Edge Cases

CI harnesses around Claude Code make agent quality measurable. Without MemU, harness scores do not compound into long-term debugging memory.

Read article

Cover for Claude Projects Keep Long Chats Scoped — But Project Context Is Not a Shared Agent Memory Layer

Earlier-version content

MemU Team

Apr 01featured

Claude Projects Keep Long Chats Scoped — But Project Context Is Not a Shared Agent Memory Layer

Claude Projects organize files and instructions for long threads. MemU adds API-addressable memory across chat, code agents, and automation.

Read article

Cover for Claude Computer Use Automates Browsers and Desktops — But Ephemeral UI State Is Not Durable Operational Memory

Earlier-version content

MemU Team

Apr 01featured

Claude Computer Use Automates Browsers and Desktops — But Ephemeral UI State Is Not Durable Operational Memory

Claude Computer Use drives real UIs for legacy systems. MemU persists selector maps and outcomes when portals redesign overnight.

Read article

Cover for CLAUDE.md Gives Claude Code a Repo Constitution — But Static Markdown Cannot Remember Yesterday's Merge Conflicts

Earlier-version content

MemU Team

Apr 01featured

CLAUDE.md Gives Claude Code a Repo Constitution — But Static Markdown Cannot Remember Yesterday's Merge Conflicts

CLAUDE.md teaches repositories to Claude Code fast. MemU remembers migrations and incidents that markdown alone never updates automatically.

Read article

Cover for Amazon Bedrock Serves Claude Inside VPCs for Regulated Teams — But Isolated Deployments Multiply Memory Silos

Earlier-version content

MemU Team

Apr 01featured

Amazon Bedrock Serves Claude Inside VPCs for Regulated Teams — But Isolated Deployments Multiply Memory Silos

Bedrock brings Anthropic Claude into private clouds. MemU offers a controllable memory plane across VPC-bound agent fleets.

Read article

Cover for Claude Code Slash Commands Standardize Team Workflows — But Macros Do Not Learn From Prior Command Outcomes

Earlier-version content

MemU Team

Apr 01featured

Claude Code Slash Commands Standardize Team Workflows — But Macros Do Not Learn From Prior Command Outcomes

Slash commands compress multi-step Claude Code flows. MemU stores failures and overrides so shortcuts get smarter over time.

Read article

Cover for Anthropic Prompt Caching Slashes Claude Latency and Cost — But Cached Prefixes Are Not Semantic Long-Term Memory

Earlier-version content

MemU Team

Apr 01featured

Anthropic Prompt Caching Slashes Claude Latency and Cost — But Cached Prefixes Are Not Semantic Long-Term Memory

Prompt caching cuts token spend on stable Claude prefixes. MemU holds facts that change weekly; caching holds text that stays stable.

Read article

Cover for Claude Parallel Tool Calls Compress Wall-Clock Agent Time — But Concurrent Results Still Evaporate Between Sessions

Earlier-version content

MemU Team

Apr 01featured

Claude Parallel Tool Calls Compress Wall-Clock Agent Time — But Concurrent Results Still Evaporate Between Sessions

Parallel tool use speeds Anthropic agent loops. MemU persists merged conclusions so the next chat does not re-fetch everything.

Read article

Cover for Claude Messages API Is the Backbone of Custom Agent Products — But Message Threads Are Not Your Memory Product

Earlier-version content

MemU Team

Apr 01featured

Claude Messages API Is the Backbone of Custom Agent Products — But Message Threads Are Not Your Memory Product

The Messages API powers most production Claude agents. MemU adds dual-mode retrieval so threads are not your only knowledge store.

Read article

Cover for Claude Code Unifies MCP, Subagents, and Terminal Tools — But Multi-Step Coding Runs Still Drop Cross-Session Facts

Earlier-version content

MemU Team

Mar 31featured

Claude Code Unifies MCP, Subagents, and Terminal Tools — But Multi-Step Coding Runs Still Drop Cross-Session Facts

Claude Code combines MCP servers, subagents, and terminal tools for agentic development. But durable AI agent memory across sessions still requires an external layer like MemU.

Read article

Loading more articles…