AI Tools Directory

Language / 语言

Claude-Mem

Claude-Mem

While a coding assistant works, it...

4.0· Editor Rating

Editor Review

What a new session usually loses is why the last change was made. Lifecycle hooks capture automatically at session start, when a prompt is submitted, after a tool runs, and when the session ends, so you do not paste context by hand each time. Layered search starts with a compact index of ids, then a timeline of what happened around them, and only then fetches full observations by id. A web viewer shows the memory stream, and sensitive content can be wrapped so it is not stored. It fits developers who open many coding sessions and still want later sessions to cite earlier decisions.

AI Tools Navigator Editorial TeamUpdated: 2026-10-09

What is Claude-Mem

Claude-Mem, by Alex Newman, is a cross-session memory plugin. While a coding assistant works, it records tool use and observations, compresses them into semantic summaries, and injects relevant context into later sessions. A local worker exposes a web viewer and search API. Observations live in a database, with vector search for hybrid retrieval. After install you can sign in for a trial of the hosted observer, use your own API key, or skip sign-in.

Basic Info

Category:

Best For

Developers working across many coding sessionsEngineers who need the reason for a past changeUsers who want memory stored locally

Difficulty: Intermediate

Claude-Mem Key Features

  • Capture through lifecycle hooks

    Observations are recorded at session start, prompt submit, after tool use, stop, and session end, without a manual export each time.

  • Three-layer memory search

    Search returns a compact index with ids, timeline shows what happened around them, and full observations are fetched in a batch by id.

  • Local viewer and database

    The worker serves a web viewer and search API. Observations and summaries sit in a local database, and a vector store handles hybrid semantic and keyword search.

  • Choice of memory provider

    After sign-in you can use the hosted observer, switch to your own API key, or skip sign-in at install time and name a provider.

Claude-Mem Key Advantages

  • Hooks record after tool use, so decisions do not have to be pasted into the next session
  • Search gives a compact index and a timeline first, and full observations are fetched by id only when needed
  • Sensitive passages can be marked for exclusion and are not written into memory
  • You can sign in for the hosted observer, use your own API key, or skip sign-in

Claude-Mem Use Cases

  • Resume a bugfix the next day

    A developer opens a new coding session, has the assistant search memory for the last bugfix notes, and continues the change.

  • Keep sensitive text out of storage

    Wrap secrets or private text with the exclusion marker so those passages are not written into memory.

  • Review observations in the viewer

    Open the web page printed when the worker starts and review automatic observations and summaries by project.

Frequently Asked Questions

▶How does Claude-Mem keep context for a later session?

Lifecycle hooks capture tool use and observations automatically, compress them into semantic summaries, and make them available to later sessions. Layered search has three steps: a compact index with ids, a timeline around an observation, then the full text fetched by id. A local worker also serves a web viewer. You do not paste the previous conversation into the new session by hand.

▶Can Claude-Mem be installed without signing in?

Yes. The installer normally asks you to sign in in the browser and offers a trial of the hosted observer. To skip that, pass a provider at install time, turn off the online opt-in, or run in a non-interactive shell so setup finishes without an account. After sign-in you can also use your own API key. If you do not subscribe when the trial ends, memory falls back to your own model plan.

▶Why are hooks missing after a global package install?

The repository says a global package install is the SDK only. It does not register plugin hooks or start the worker. To have memory show up in a new session, use the install command in the project docs or install from the assistant's plugin marketplace. Restart the assistant afterward. A recent Node.js is required. Bun and the tool used for vector search are installed automatically if they are missing.

Latest News

View More

User Reviews

Real reviews and feedback from users

Related Tools

Seele AI

Seele AI

+3

Seele AI is an AI-powered game development tool th...

Skywork

Skywork

Skywork is Kunlun Tech's Super Agent platform buil...

Flowith

Flowith

+4

Flowith AI is a smart agent workspace built around...

Manus

Manus

+4

Manus is an action engine that goes beyond providi...

Operator

Operator

Operator is OpenAI's first AI agent — an AI tool t...

金灵AI

金灵AI

+3

Gilin AI is a professional financial research AI a...

星辰Agent

星辰Agent

+4

Xunfei Star Agent is an AI agent development platf...

MasterAgent

MasterAgent

MasterAgent is the world's first L4-level multi-ag...