Persistent memory for any AI agent

Your agents save what matters and find it instantly — across every machine, every session, every AI tool you use. Encrypted on your machine before it leaves; we never see what you save. Free to start, no card required.

Works with OpenClaw, Claude Code, Cursor, Aider, Codex CLI, Gemini CLI, Goose, and more — see all integrations.

One command — installs in memory mode and runs setup.

curl -fsSL /install.sh | MEMORYCLAW_MODE=memory bash && memoryclaw setup

Auto-detects Claude Code, Cursor, Codex CLI, Gemini CLI, Aider, Goose, OpenClaw, and Claude Desktop. Pick which to wire up in the setup picker — see memoryclaw.ai/setup for per-agent guides + Cline / Continue.

Installs in seconds. Sign in, pick a passphrase, and your agent has memory.

Built for developers who care

Unlimited

Agents per account

Zero-knowledge

We never see your data

AES-256

Encrypted on your machine

Command install

Get started in three steps

From install to your first memory in under a minute. No config files, no setup wizard, no friction.

Install

One command sets MemoryClaw up for any AI agent — Claude Code, Cursor, Aider, OpenClaw, and more. Works on macOS, Linux, Windows, and WSL.

Connect

The installer signs you in automatically — your browser opens, you create an account, and your Neural Memory is ready.

Save & Recall

Your agents save what matters and find it later, by meaning. Threads keep separate projects separate when you want them to. Everything is encrypted on your machine before it leaves.

Long-term memory complements long-context

Long context windows are great for one-shot reasoning over a big input. Persistent context memory is for the recurring facts you don't want to re-load every turn — saved once, recallable across every session, every machine, every agent you connect.

How it grows

Memory that evolves

facts link, merge, and update

Each context memory captures one thing — a fact, a preference, a result. As you work, related memories link together, repeated patterns merge automatically, and newer information overrides older when they conflict. Encrypted on your machine before any of it leaves.

Speed of recall

Use it mid-turn

single network round-trip

The agent searches by meaning, not by exact words — and gets results fast enough to slot into its existing reasoning loop. No re-reading prior turns, no scrolling back through long context.

Where it lives

Across everything

machines, sessions, agents

Saved once on your laptop, recallable tonight on your desktop or after a fresh install — and shared across every AI agent you connect. Still there mid-flow during long sessions when the context window would otherwise roll off and the agent would forget what it just learned.

Free tier includes Neural Memory; encrypted backup is available on every tier and opt-in per agent. No card required. See current quotas on the pricing page or read the deep dive in our long-context vs Neural Memory post.

Two pillars no other hosted memory matches

Every tier — including Free — gets both. They aren't add-ons and they aren't marketing fluff: they're how MemoryClaw works at the protocol level.

Zero-knowledge encryption

Encrypted on your machine. Your passphrase derives the key locally; the server never sees plaintext. We can't read your memories, can't hand them to law enforcement under subpoena, can't lose them in a breach. The trade-off is honest: lose the passphrase, lose the data.

Memory that evolves

Memories that connect — not a flat pile of files. Each fact links to related facts as you work. When patterns repeat, similar memories merge automatically. When new information contradicts an older fact, the memory updates itself instead of leaving the agent confused. The agent gets a structured map back, not a bag of similar strings.

Everything you need, nothing you don't

Encrypted memory, threaded contexts, smart recall, and optional encrypted backup. Built for AI developers.

Encrypted on your machine

Your data is encrypted with AES-256-GCM on your laptop before it ever leaves. The server holds ciphertext, never plaintext — we couldn't read your memories if we tried.

Neural Memory

Your agents save what matters, find it by meaning, and link related facts as they work. A structured map of what they know — not a flat pile of files.

Memory Threads

Keep different projects in separate threads, or share memory across all your agents — you decide the scope per thread. Switch contexts without losing what you already know.

Smart Recall

Find anything by meaning, not just exact words. Your agent searches its memory the way you'd describe it — natural language, not exact strings.

Works with your agent

Claude Code, Cursor, Aider, Codex CLI, Gemini CLI, Goose, OpenClaw, Cline, Continue — one CLI, every agent. macOS, Linux, Windows (WSL).

Encrypted Backup · for any agent

Encrypted backup of your Claude Code, Cursor, Codex, OpenClaw setup — skills, settings, credentials. Pick one or all. Zero-knowledge: encrypted on your machine before it leaves.