Retrieve the Right Memory Before the Agent Answers

One query embedding, readable results, and no LLM in the retrieval path.

The host adapter retrieves memory before a task. memU embeds the query once, ranks searchable file segments and workspace resources by vector similarity, then returns the matching segments, complete recall files, and resources to the agent.

Install memU

How Retrieval Works

Embed Once

A natural-language query is embedded once and reused across the configured retrieval layers.

Rank File Segments

Searchable slices of memory and skill files are ranked by cosine similarity to find semantically relevant text.

Roll Up Complete Files

Matching segments point back to their complete recall files, so the agent receives readable context instead of isolated fragments.

Retrieve Workspace Resources

Workspace resource records are ranked alongside file segments and returned as a separate result layer.

What Agents Can Retrieve

Project Decisions

Bring durable architecture choices, constraints, and project conventions back into a new session.

Reusable Skills

Recall workflows distilled from earlier sessions when a similar task, failure mode, or tool appears again.

Workspace Context

Surface relevant indexed resource records from the workspace without asking the user to repeat the background.

Cross-Agent Continuity

Let one connected agent retrieve useful memory or skills learned from another host using the same backend and scope.

User Preferences

Restore durable preferences that the agent previously chose to preserve as readable memory.

Prior Outcomes

Bring back tested approaches, edge cases, and pitfalls before the agent repeats the same work.

Designed for an Agent’s Critical Path

No Retrieval LLM

Retrieval performs embedding and vector ranking only; there is no query rewrite, sufficiency check, or summarization call.

Readable Context

Segment hits roll up to complete memory or skill files that the host agent can inspect directly.

Track-Aware

Memory and skill files retain their track, allowing callers to preserve the role of each kind of recall file.

Explicit Scope

Retrieval is constrained by the configured user and agent scope before results are ranked.

Ready to Unlock the Full Potential of AI Memory?

Memory Storage

Store readable memory and skill files with searchable segments in a shared Cloud, SQLite, or Postgres backend.

Explore Memory Storage

Memory Retrieval

Embed a query once, rank relevant file segments and workspace resources, then return the complete readable files to the agent.

Now Here

Automatic Skill Extraction

Let the agent turn useful session history into reusable Markdown skills, then index those skills for future tasks.

Explore Skill Extraction

Cross-Agent Memory

Connect multiple host adapters to one configured backend so skills learned in one agent can be retrieved by another.

Explore Cross-Agent Memory

Agent-Driven Memory

Keep judgment and synthesis inside the agent while memU handles durable storage, embeddings, and retrieval.

Explore Agent-Driven Memory

Readable Markdown

Give agents inspectable memory and skill files instead of hiding all useful context inside opaque vector records.

Explore Markdown Memory

Coding Skills

Capture coding sessions and compound successful workflows, debugging lessons, and project conventions into reusable skills.

Explore Coding Skills

Open Source

Inspect the lightweight Python implementation, run locally, and adapt host integrations for the agents you already use.

Explore Open Source

Connect Your Agents to memU

Cloud Platform

Get a memU Cloud key, then give the installation prompt to your agent. The skill configures its host adapter against one managed, cross-device memory backend.

Get the Install Prompt

Self-Hosted Open Source

Ask your agent to follow the open-source SKILL.md. It installs the CLI, configures SQLite or Postgres and an embedding provider, then verifies retrieval with the host adapter.

View SKILL.md

Add Another Agent

Run the installation skill in each additional host. Every configured adapter can record its own sessions and retrieve from the same shared backend.

See Host Adapters

FAQ

memU is a lightweight, agent-driven memory system for the agents you already use. It gives connected agents one shared, inspectable memory layer across sessions, agents, and devices.

The memU installation skill configures a host adapter for your agent. A scheduled task captures new session history, the agent decides what memory or skill Markdown to write, and the adapter retrieves relevant files before future tasks. Users do not need to integrate a public memory API.

No. memU's memory service makes no LLM or chat calls. Judgment and synthesis remain inside the host agent; memU stores the Markdown it prepared, creates embeddings, and retrieves relevant files.

The query is embedded once. memU ranks searchable file segments and workspace resources by vector similarity, rolls matching segments up to their complete recall files, and returns segments, files, and resources without an LLM in the retrieval path.

memU includes dedicated adapters for Codex, Claude Code, Cursor, OpenClaw, Hermes, and WorkBuddy. A generic adapter can detect compatible session logs and instruction files for other agents.

Host adapters use the same configured Cloud or local backend. When they use the same memory scope, a skill learned from one host's sessions can be retrieved by another connected host.

Yes. The self-hosted mode supports SQLite for local use and Postgres with pgvector for larger or concurrent stores. Local mode requires a configured embedding provider. memU Cloud provides a managed, cross-device backend.

Yes. The Python package, CLI, host adapters, and core memory service are available on GitHub under the Apache-2.0 license.

Retrieve Shared Memory in the Agents You Already Use

Install the host adapter, connect it to your shared backend, and let the agent retrieve relevant memory before future tasks.

Install memU