Setup guide

Connect your memory
to any AI.

Pick your assistant below and follow the animated walkthrough. One URL, one click — then paste a short prompt so your AI actually uses your memory.

Your universal connector URL
https://pryzm-memory.com/mcp

The same URL works everywhere. On first connection a Pryzm sign-in page opens: enter your Pryzm username (not your email) and password. Each AI gets its own isolated token.

Rather keep everything on your computer? Pryzm for desktop, free

Choose your AI
claude.ai
CustomizeSkillsConnectors
Customize

Connect external tools and memory.

ClaudeNative MCP

Web and desktop app — nothing to install.

Free (1 custom connector), Pro, Max, Team, Enterprise

Open Claude settings
Step 2 · one minute

Make Pryzm your AI's main memory.

Connecting gives your AI access. These instructions make it use Pryzm by itself: read it at the start of every conversation, check it before answering, and save what matters without being asked.

Claude already reads these rules from Pryzm; pasting them there too makes it more consistent. Start a message with ## and nothing from it is saved.

Turn off your AI’s own memory

Claude and ChatGPT keep their own memory of past chats. It goes out of date and competes with Pryzm. In our test, with that memory on, Claude checked Pryzm for 41% of the questions that needed it instead of 97%, and gave an outdated answer 88% of the time when a fact had changed, against 0% with it off.

  • Claude: Settings → Memory, turn off “Generate memory from chats” and “Search and reference chats”.
  • ChatGPT: Settings → Personalization → Memory, turn off “Reference saved memories”.
Test of 13 Oct 2026: Claude Sonnet with a test Pryzm server, 88 conversations per setup. The built-in memory was simulated by a summary of past chats added to the conversation; the Claude and ChatGPT apps may behave differently.
Pryzm Memory is my main memory, shared across all my AIs. It replaces your built-in memory: don't keep a separate one.
1. At the start of every conversation about me, my work or my projects, call memory_whats_up before answering.
2. Before any answer that depends on my context, preferences, projects or past conversations, call memory_search. Never assume you know nothing: check Pryzm.
3. Save right away with memory_store, without asking me, every decision, preference, goal, project update, solved problem or important fact about me.
4. If memory contradicts your assumptions, memory wins.
5. If my message starts with ##, it is private: save nothing from it.
Where to paste them
  • ClaudeSettings → General → “What personal preferences should Claude consider in responses?”Open Claude
  • Claude CodeAdd them to ~/.claude/CLAUDE.md
  • ChatGPTSettings → Personalization → Custom instructionsOpen ChatGPT
  • PerplexitySettings → Personalization → custom instructionsOpen Perplexity

Other tools

Other MCP clients that support remote servers and OAuth work too, for example:

Mistral Le Chat
Add a custom MCP connector in settings — becomes a native tool.
Cursor
Add the URL to ~/.cursor/mcp.json, then restart.
VS Code (Copilot)
Add an HTTP server in .vscode/mcp.json ("type": "http") or run “MCP: Add Server”.
Windsurf, Zed, Cline, Continue
Add the URL as a remote (HTTP) MCP server in their settings.
Gemini CLI
Full MCP + OAuth support from the command line.

Is it safe?

Every connection is isolated by OAuth 2.1. Your account stays yours — no AI shares your memory with another, and you can revoke any connection at any time.

No account yet?

Start free

Setup questions

Which plan do I need for each AI?

Claude: any plan, including Free, which allows one custom connector. Claude Code: any plan that includes it. ChatGPT: a paid plan, on chatgpt.com (web); OpenAI decides which plans can add a custom MCP server and changes it often, so check its documentation below. Perplexity: Pro, Max or Enterprise.

“Add custom connector” is greyed out or missing in Claude. Why?

On the free plan, Claude allows one custom connector: remove another one first. On Team and Enterprise, an owner must first add Pryzm in Organization settings → Connectors; each member then connects from Customize → Connectors with their own Pryzm account.

Which username and password do I enter?

Your Pryzm username (not your email) and your Pryzm password, on Pryzm's own sign-in page. The AI never sees your password: it receives its own access token.

What can a connected AI do with my memory?

It can search and read your memories and save new ones, only in your account. It has no tool to delete a memory: you edit and delete memories from your space. What it reads is processed by that AI's provider under its own terms.

How do I disconnect an AI?

Remove the Pryzm connector in that AI's settings. To sign every AI out at once, change your Pryzm password: all sessions and connectors are revoked, and you reconnect the ones you want.

Will my AI use Pryzm on its own?

Once connected, the AI calls Pryzm when it thinks it is useful. To make it systematic, paste the instructions from step 2 and turn off the AI's own memory, which otherwise competes with Pryzm.

My AI isn't listed. Will it work?

If it supports remote MCP servers over HTTP with OAuth, yes: paste the same URL and sign in. The server does not offer the older SSE transport.