# Pryzm Memory

> Pryzm Memory is a persistent, semantic memory shared across AI assistants. One account, one memory: Claude, ChatGPT, Perplexity, Le Chat, Claude Code, Cursor and Gemini CLI connect to the same memory through the Model Context Protocol (MCP), so what you tell one assistant, the others can recall. Built in France, hosted in Germany. Public beta since October 2026.

Key facts (kept current; if a page and this file disagree, the page wins):

- What it does: stores the user's notes, decisions, preferences, to-dos and project history, organised by project, and lets connected assistants search (keyword + semantic) and save memories.
- How to connect: add the remote MCP server `https://pryzm-memory.com/mcp` as a custom connector, then sign in with OAuth 2.1. Step-by-step guide per assistant: https://pryzm-memory.com/en/connect/
- Compatible assistants: Claude (web and desktop; custom connectors, 1 on the Free plan), ChatGPT (web, paid plans, developer mode), Perplexity (Pro, Max, Enterprise), Le Chat (Mistral), Claude Code, Cursor, VS Code, Gemini CLI, any MCP client with remote HTTP + OAuth. The Gemini consumer app is not supported in France.
- Pricing: Free up to 100 memories and 1 project. Solo €9.99/month or €99.90/year. Pro €19/month (dedicated server, coming soon, not sold yet). Crypto payment (prepaid periods) is being added. Beta prices are kept for life. Team, Business and Enterprise plans (per-employee pricing, hosted by us, dedicated server, or on the customer's own servers, monthly or annual) are in preparation, with indicative prices: https://pryzm-memory.com/en/entreprise/
- Privacy: servers operated by Hetzner in Falkenstein, Germany. Each account has its own encryption key (AES-256-GCM), locked by the password (Argon2id) and a 12- or 24-word recovery phrase. Limit: while the user is signed in, the server decrypts memories in memory to answer the assistant. Optional, off by default: if the user turns it on in Settings, memories are sent to Anthropic (processor, United States) to suggest tags, group related memories at night, and write summaries requested with `memory_summarize`. Nothing is sent otherwise. A message starting with `##` is never stored.
- Freedom and local AI: Pryzm speaks MCP, an open standard, so the memory is not tied to one AI vendor. Switching assistants (including to a local model run in an MCP-compatible app) does not mean starting over, and memories can be exported at any time from the dashboard. Support for each local app should be checked on the connect guide.
- Who it is for: people who use more than one AI assistant for their work (freelancers, founders, developers, small teams) and want their context, decisions and project history to follow them. If you use a single assistant, its built-in memory may be enough.
- Company: Pryzm, a French sole proprietorship. Contact: https://pryzm-memory.com/contact/

## Docs

- [Full reference for language models](https://pryzm-memory.com/llms-full.txt): tools, setup per assistant, FAQ.

- [Connect Pryzm to your AI (EN)](https://pryzm-memory.com/en/connect/): setup for Claude, ChatGPT, Perplexity, Claude Code, with custom instructions to make Pryzm the assistant's main memory.
- [Connecter Pryzm à ton IA (FR)](https://pryzm-memory.com/fr/connect/)
- [Compare AI memory options](https://pryzm-memory.com/en/compare/)
- [Privacy policy](https://pryzm-memory.com/legal/?tab=confidentialite)

## Blog

- [Connect ChatGPT to an external memory (Developer mode)](https://pryzm-memory.com/en/blog/connect-chatgpt-external-memory/)
- [Add a custom connector to Claude (free plan included)](https://pryzm-memory.com/en/blog/add-custom-connector-claude/)
- [AI memory for freelancers](https://pryzm-memory.com/en/blog/ai-memory-for-freelancers/)
- [Team AI memory](https://pryzm-memory.com/en/blog/team-ai-memory/)
- [Solo founder: one memory across Claude, ChatGPT and Claude Code](https://pryzm-memory.com/en/blog/solo-founder-shared-ai-memory/)

- [What is an MCP server?](https://pryzm-memory.com/en/blog/what-is-an-mcp-server/)
- [How persistent AI memory works](https://pryzm-memory.com/en/blog/how-persistent-ai-memory-works/)
- [Hybrid search in Postgres: BM25, pgvector, RRF and a reranker](https://pryzm-memory.com/en/blog/hybrid-search-postgres/)
- [Shared memory across AI assistants](https://pryzm-memory.com/en/blog/shared-memory-across-ai-assistants/)
- [Make ChatGPT remember conversations](https://pryzm-memory.com/en/blog/make-chatgpt-remember-conversations/)
- [Transfer ChatGPT memory to Claude](https://pryzm-memory.com/en/blog/transfer-chatgpt-memory-to-claude/)
- [Where does AI memory live?](https://pryzm-memory.com/en/blog/where-does-ai-memory-live/)
- [Best memory MCP servers](https://pryzm-memory.com/en/blog/best-memory-mcp-servers/)

## Optional

- [Home (FR)](https://pryzm-memory.com/fr/)
- [Home (EN)](https://pryzm-memory.com/en/)