Best Memory Tools for Claude Code (2026)

Engram is a single Go binary with SQLite full text search behind an MCP server, a CLI, an HTTP API and a terminal UI. The agent saves decisions and context per project and recalls them in the next session. It detects the project from the repo, so memories from one codebase do not leak into another, and engram tui lets you browse what the agent has stored, which I find more reassuring than a black box.

brew install gentleman-programming/tap/engram
engram setup claude-code

Setup installs a Claude Code plugin with hooks; restart Claude Code afterwards. The same engram setup command supports Codex, Cursor, Gemini CLI, OpenCode and about ten others. The caveat: search is full text, not semantic, so memories need the right words to come back. Cloud sync exists but is optional, and it is MIT licensed.

ai-memory by Fabio Akita targets the exact pain of quitting Claude Code mid-task and opening Codex in the same folder. The second agent picks up a handoff with where you left off, what failed and what is still open. It works across more than twenty harnesses and stores memory as plain markdown in a git-backed wiki, so you can read and edit it yourself.

curl -fsSL https://github.com/akitaonrails/ai-memory/releases/latest/download/install.sh | sh
ai-memory setup

The installer checks the release checksum, and ai-memory setup shows a plan of every agent it will wire before it changes anything. Claude Code users can also install it as a plugin: claude plugin marketplace add akitaonrails/ai-memory && claude plugin install ai-memory@ai-memory. The caveat: it runs a background per-user service, which is more moving parts than a plain MCP server. It is MIT licensed with no paid tier.

Graphicle combines two things most people run separately: a code graph of the repo (calls, imports, impact) and a memory of docs, decisions and past agent sessions. Every excerpt it returns cites the exact file and lines, and memories the agent writes wait for your approval before they count. It runs locally with no account, no API key and no telemetry.

brew install graphicle-ai/tap/graphicle
cd your-repo
graphicle setup

Setup indexes the repo, connects the agents it finds, turns on session capture and installs Git hooks to keep the index current. It lists and backs up every file it changes, and graphicle disconnect undoes all of it. The caveat: it touches your repo (hooks and an AGENTS.md section unless Claude Code is the only agent), so read the change list before you accept.

Beads is memory shaped like an issue tracker. Instead of a loose markdown TODO file, the agent works from a dependency-aware task graph: it asks what is ready, claims a task, closes it and records durable facts. When the context compacts halfway through a big job, the plan is still there, and several agents can share one backlog without stepping on each other.

brew install beads
cd your-project
bd init
bd setup claude

There is also npm install -g @beads/bd if you prefer npm. The caveat: it remembers work, not general knowledge. It will not recall why you picked Postgres last month unless that ended up in a task, so it pairs well with one of the other tools here.

Mem0 is the best known memory layer for AI apps, and it now ships a Claude Code plugin with hooks that recall memories on your first prompt and extract new ones in the background. Search is semantic, so a memory about "payment retries" can come back when you ask about "failed charges".

claude plugin marketplace add mem0ai/mem0
claude plugin install mem0@mem0-plugins --scope user --config api_key="$MEM0_API_KEY"

The caveats: the plugin uses the hosted platform, so your project notes leave your machine, and you need an API key. The Hobby tier is free with 10,000 adds and 1,000 retrievals a month; Starter is $19 a month. The open source library can be self-hosted, but that is more work than any other option here.