AI memory is the persistent context an AI assistant keeps about you between conversations: your preferences, projects, and facts it has learned, so every session starts already knowing you. Native memory lives inside one product; MemoryRouter provides portable memory that works across the AI tools you use.

Yes. You approve everything before it saves, your vault is encrypted at rest and in transit and isolated to your account, it is never shared or trained on, and you can delete it anytime.

One memory for every AI

Q: What is the difference between saved memory and chat history?

Chat history is the raw log of past conversations inside one tool. Saved memory is the distilled, reusable context an assistant carries forward, like your role, your goals, and how you like to work. MemoryRouter lets you review and approve what becomes saved memory, then makes that memory available across every connected assistant.

Q: Can I share memory between ChatGPT and Claude?

Yes. That is the core of MemoryRouter. You transfer what one assistant already knows into your own vault, then connect Claude, Cowork, coding agents, and other tools to the same memory, so context you gave in one shows up in the others.

Q: Is this the same as long-term memory for an LLM?

For developers, yes in practice. MemoryRouter exposes a hosted MCP endpoint and a memory API so your agents get persistent, long-term memory across sessions instead of forgetting everything when a context window ends.

Q: How much does MemoryRouter cost?

MemoryRouter is $20 a month after a 14-day free trial, with no charge today and cancel anytime. Every plan includes 200M memory tokens each month; additional usage is $0.50 per 1M tokens. Companies that need shared memory can explore Enterprise.

AI memory is the persistent context an assistant keeps about you between conversations. Most tools keep their own. MemoryRouter gives you one cross-AI memory that you own and connect to every assistant, so the context you give in one shows up in all of them.

Teach it once, then use ChatGPT, Claude, Cowork, Claude Code, Codex, and OpenClaw without re-explaining who you are or what you are working on. Your memory travels with you, not with a single model vendor.

Start Basic transfer Read developer docs

Quickstart: your first API request

MemoryRouter speaks the OpenAI API. Point a request at api.memoryrouter.ai, keep your own provider key, and the memory layer takes care of recall and storage on every call. About two minutes from here to your first response.

1. Get a Memory Key
Sign up and mint a Memory Key. It starts with mk_ and it points to one memory vault.
Get a Memory Key

2. Send one request

Copy this into a terminal, drop in your Memory Key and your OpenAI key, and run it.

terminal

curl -X POST https://api.memoryrouter.ai/v1/chat/completions \
  -H "X-Memory-Key: mk_xxxxxxxxxxxxxxxx" \
  -H "Authorization: Bearer sk-your-openai-key" \
  -H "Content-Type: application/json" \
  -d '{"model": "openai/gpt-5.5", "messages": [{"role": "user", "content": "Remember that I train at 6am."}]}'

Using the OpenAI SDK in Python? Point it at MemoryRouter and add one header. Everything else stays the same:

python

from openai import OpenAI

client = OpenAI(
    api_key="sk-your-openai-key",  # Goes to provider
    base_url="https://api.memoryrouter.ai/v1",
    default_headers={
        "X-Memory-Key": "mk_xxxxxxxxxxxxxxxx"  # MemoryRouter auth
    }
)

3. Ask it something tomorrow
Ask it something today. Ask it something tomorrow. It remembers. Full reference: docs.memoryrouter.ai/quickstart.

What “AI memory” actually means

When people say AI memory, persistent AI memory, or long-term memory for an LLM, they mean the same underlying idea: an assistant that remembers useful context about you across sessions instead of starting from a blank slate every time. That context can be your role, your projects, your writing style, or facts you have already explained.

The open question is where that memory lives. If it lives inside one product, it only helps you inside that product. If it lives in a vault you own, it can follow you everywhere.

Native memory versus portable memory

Native memory (built into one tool)

ChatGPT, Claude, and others each keep their own memory. It is convenient, but it stays locked inside that one product. Switch to a different assistant, tool, or model and you start over.

Portable memory (MemoryRouter)

One vault that you own, connected to every AI you use. Teach it once and the same context is available in ChatGPT, Claude, Cowork, Claude Code, Codex, and OpenClaw. Your memory is yours, not any single model vendor’s.

Saved memory versus chat history

Chat history is the raw log of everything you have typed in one tool. Saved memory is the distilled, reusable part worth carrying forward, like your goals and preferences. MemoryRouter lets you review and approve what becomes saved memory, then makes that approved memory available across every connected assistant.

The fastest way to see it is the Basic transfer: MemoryRouter gives you one prompt to paste into ChatGPT, ChatGPT returns everything it knows about you, and you approve exactly what saves. Read the Basic transfer walkthrough.

Want more than the current summary? The Advanced importer reads your official ChatGPT export, rebuilds the conversation graph, and lets you preview and approve exactly what imports before anything saves. Read the Advanced import guide.

Who cross-AI memory is for

For individuals

Stop re-explaining who you are, what you are building, and how you like to work every time you open a new tool. Your preferences, projects, and history follow you across assistants.

What is MemoryRouter?

For teams

Shared memory means a team can build on a common base of context instead of re-briefing every assistant. Companies that need shared memory can explore Enterprise.

Explore Enterprise

For developers

A hosted MCP endpoint and a memory API let you give your own agents persistent, low-latency, long-term memory across sessions, with OAuth-bound access to each user’s vault.

See the MCP memory server

How the architecture works: one memory, every AI

Your memory lives in a single vault. Each assistant connects to it, either through a native connector, automatic hooks, or the hosted MCP endpoint. When an assistant needs context, it reads from your vault; when you approve something new, it writes back to the same vault.

ChatGPT

Add MemoryRouter as a custom connector and your context follows every piece of work you do in ChatGPT.

Claude and Cowork

Add the MemoryRouter connector so Claude and Cowork use the same personal memory.

Claude Code

Hooks retrieve and capture context automatically, so a new coding session starts with your project loaded.

Codex

OpenAI Codex gets the same automatic hook-based memory, so a fresh session begins with your context in place.

OpenClaw

Installed as the mr-memory plugin, MemoryRouter gives your own agent workflows persistent, low-latency memory.

MCP memory server

One hosted MCP endpoint connects Claude, ChatGPT, coding agents, and other compatible clients to your vault.

Your memory, under your control

You approve what gets remembered

Nothing saves to your memory until you have reviewed and approved it.

Encrypted and isolated to your account

Your vault is encrypted at rest and in transit, isolated to your account, and never shared or trained on.

Delete anytime

Wipe your entire memory vault whenever you want. No lock-in, no waiting period. Your data is yours.

Choose a path

Most people start with the Basic transfer and connect one tool, then add the others over time. MemoryRouter is $20 a month after a 14-day free trial, with no charge today and cancel anytime. Every plan includes 200M memory tokens each month; additional usage is $0.50 per 1M tokens.

Individuals: learn what MemoryRouter is, move an assistant across with ChatGPT to Claude memory transfer, or see full pricing.
Developers: connect the MCP memory server or read the developer docs.
Teams: explore Enterprise for shared memory.

Related guides

Deeper reading on agent memory, memory APIs, and running one memory across your tools:

Persistent memory for AI agents Evaluating a long-term memory API What an AI agent memory layer does Vector memory vs RAG for agents Chatbot memory across sessions A ChatGPT memory alternative Claude memory across sessions Beyond CLAUDE.md in Claude Code Cursor rules vs memory One shared memory across AI tools Which coding tools remember across sessions Memory guides for every AI tool Fix lost context and memory problems Honest memory comparisons Does Codex have memory?Does Cursor remember past conversations?

AI memory FAQ

What is AI memory?

What is the difference between saved memory and chat history?

Can I share memory between ChatGPT and Claude?

Is this the same as long-term memory for an LLM?

How much does MemoryRouter cost?

Is my data private?