Yes. The hooks run at deterministic Claude Code lifecycle points: SessionStart, UserPromptSubmit, Stop, PreCompact, PostCompact, resume, and fork. Recall and capture happen every turn whether or not Claude decides to call a tool.

npx -y memoryrouter-claude off removes the hooks. npx -y memoryrouter-claude off --purge also deletes the private key and local state. Everything else in your Claude Code setup stays untouched.

Claude Code

Claude Code loses your project every time it compacts
One command fixes that.

Q: What does MemoryRouter give Claude Code?

A memory that outlives the context window. Hooks recall the relevant project context before every prompt and save every completed turn. Open a new session tomorrow, on this machine or another, and Claude Code already knows what you decided today.

Q: Does it survive compaction?

Yes. That is the whole point. Right after compaction, MemoryRouter re-retrieves your project memory and injects it, so Claude keeps going like nothing happened. Same for resume and fork.

Q: What exactly gets saved?

Your prompt and Claude’s final, user-visible response for each completed turn. Tool calls, patches, intermediate assistant text, thinking, compaction summaries, and subagent chatter are excluded. Your memory stays clean and useful.

Q: Is memory scoped to my project?

Yes. Every project gets its own partition. Worktrees of the same git repository share one memory; unrelated projects never mix. Clone the repo on a new machine and your project memory is waiting.

Q: Where does my Memory Key live?

In a private file under ~/.memoryrouter/ with owner-only permissions. It is never written into .claude settings, .mcp.json, or your repository, so the hook settings are safe to commit and share with your team.

Q: Do I need an Anthropic API key?

No. Your existing Claude Code account is all you need. MemoryRouter never touches your inference.

Q: Can Claude search memory on demand too?

Yes. Run npx -y memoryrouter-claude mcp install --local then claude mcp login memoryrouter and Claude gets explicit search_memories and store_memory tools for the moments you want to say "remember this" or "what did we decide about X".

Q: Is the same memory available in Codex, ChatGPT, and Claude?

Yes. One MemoryRouter vault, every AI you use. Decide something in Claude Code and Codex knows it. Explain your project to ChatGPT once and Claude Code starts with it.

Your architecture. Your decisions. The bug you spent three hours on together. All of it, in every session, on every machine.

One command. Permanently.

See the difference

AFTER COMPACTION

You: Keep going with the plan

Claude: I no longer have the earlier context. Could you restate the plan and which files we were changing?

An afternoon of context, gone in one compaction.

WITH MEMORYROUTER

You: Keep going with the plan

Claude: Continuing: step 3 of the queue migration, swapping the Redis client behind the adapter we agreed on. Starting with lib/queue/adapter.ts.

Memory re-injected the instant compaction ends.

Sound familiar?

Compaction wiping out hours of built-up context

Re-explaining the codebase in every new session

Decisions from last week that Claude has no idea about

With MemoryRouter

Relevant project memory injected before every prompt

Every completed turn saved: your prompt and Claude's final answer. Zero tool noise.

Memory restored automatically after compaction, resume, and fork

Set up in 3 minutes

Grab your memory key, open a terminal in your project, run one command.

Step 1: Install the hooks

$ npx -y memoryrouter-claude init

The installer prompts for your Memory Key, validates it, and asks where to install: this project, this checkout only, or every project on this computer. The key stays in your home folder, never in the repo.

Step 2: Verify

$ npx -y memoryrouter-claude doctor

Start a new Claude Code session, then run doctor. A healthy setup ends with "Automatic memory integration is healthy."

Step 3: Watch it click

Tell Claude one unique fact about the project. Finish the turn. Open a brand new session in the same project and ask for it. It is already there.

Just tell Claude Code to do it

Paste this into a Claude Code session. It asks for your key, runs the install, and verifies everything.

Install MemoryRouter automatic memory hooks for me in this project.

1. Ask me for my MemoryRouter Memory Key (from app.memoryrouter.ai) and wait for my reply before running anything.
2. Check prerequisites: run `node --version`. I need Node 18 or newer.
3. Run: npx -y memoryrouter-claude init
   When the installer prompts for the key, use the one I gave you. Do not write the key into any file in this repository.
4. Run: npx -y memoryrouter-claude doctor
   Report the result in plain language. A healthy setup ends with "Automatic memory integration is healthy."
5. Tell me to start a new Claude Code session in this project, tell you one unique fact, then open another new session and ask for that fact. That is how I confirm memory is working.

Never print, log, or repeat my Memory Key in plain text after I give it to you.

How it works

Without MemoryRouter:

You Claude Code Model Response (gone at compaction)

With MemoryRouter:

You Claude Code (memory recalled on UserPromptSubmit) Model Response (saved on Stop)

Hooks fire at deterministic Claude Code lifecycle points: SessionStart warms the vault, UserPromptSubmit recalls and captures, Stop saves the final answer, and compaction, resume, and fork all re-inject memory. Your Claude account and inference stay exactly where they are. MemoryRouter only stores and retrieves memories.

Pick your scope

Project

./.claude/settings.json

Best for teams. Commit the hook settings and every teammate gets memory. Keys stay private on each machine.

Local

./.claude/settings.local.json

Just you, just this checkout.

User

~/.claude/settings.json

Every project on this computer, each with its own project memory.

Project-scoped

Each project gets its own memory partition. Worktrees of the same repo share it. Unrelated projects never mix.

Clean capture

Your prompt and Claude's final answer. Tool calls, patches, thinking, and subagent chatter are excluded automatically.

Survives everything

Compaction, resume, fork, new sessions, new machines, Claude Code updates. Your memory persists through all of it.

Key stays private

Your Memory Key lives under ~/.memoryrouter/ with owner-only permissions. Never in your repo or Claude settings.

Want Claude to search memory on demand too?

Hooks are the automatic engine. Add the MCP connection and Claude also gets explicit search and store tools for when you want to ask directly.

Add explicit memory tools

$ npx -y memoryrouter-claude mcp install --local

$ claude mcp login memoryrouter

All commands

npx -y memoryrouter-claude init # Install or upgrade hooks

npx -y memoryrouter-claude doctor # Health check

npx -y memoryrouter-claude status # Vault stats and project scope

npx -y memoryrouter-claude mcp install --local # Explicit MCP tools

npx -y memoryrouter-claude off # Remove hooks

npx -y memoryrouter-claude off --purge # Remove hooks and local state

Start free. Then just $20/mo.

14-day free trial. Cancel anytime.

14-DAY FREE TRIAL

$20/mo

200M tokens included every month

200M memory tokens / mo included
Unlimited projects
Same memory in ChatGPT, Claude, Codex, and OpenClaw
Sub-50ms retrieval
Your keys, your data, your control

Start free

No inference markup, ever.

FAQ

What does MemoryRouter give Claude Code?

Is this automatic?

Does it survive compaction?

What exactly gets saved?

Is memory scoped to my project?

Where does my Memory Key live?

Do I need an Anthropic API key?

Can Claude search memory on demand too?

Is the same memory available in Codex, ChatGPT, and Claude?

How do I remove it?

Never lose a session to compaction again

14 days free. Three minutes to install. Tomorrow's first session already knows everything from today.

Get your memory key Compare memory behavior

Full Claude Code guide npm package Also using Codex?