How to Give Any AI Your Full Context with One Link
Short answer: put your conversation history from every assistant into Memex, create a Memex link, and paste that link into whichever AI you are using. The link serves your archive as clean, searchable Markdown. The assistant searches it, opens the conversations that matter, and answers with your real context instead of starting from zero.
The problem: every AI starts cold
Ask a fresh chatbot "what should I work on next?" and it will ask you who you are. It does not know about the plan you made last week, the constraints you already discovered, or the decisions you already took. That knowledge exists, but it is scattered across ChatGPT, Claude, Gemini and whatever else you used, and none of them share.
The usual workarounds do not scale:
- Re-explaining costs time every single session and is always incomplete.
- Pasting old chats runs into context limits and depends on you remembering which chats matter.
- Built-in memory only works inside one product and only keeps a summary.
The idea: context as a link
A Memex link is a single URL that represents your archive. When an assistant fetches it, it gets:
- An index of your conversations as a Markdown table, with dates, titles, summaries and keywords, plus search and filter parameters.
- Each conversation as its own Markdown page, in order, with a stable ID.
- Instructions for agents, including a machine-readable protocol and an OpenAPI description, so the assistant knows how to search and what it is allowed to do.
The assistant does not load your whole history into its context window. It searches first, then opens only the conversations relevant to your question. That is why it keeps working as your archive grows into thousands of conversations.
How to use it
- Build your archive. Upload exports from ChatGPT, Claude, Gemini, DeepSeek or WhatsApp.
- Copy your Memex link from the dashboard. Memex also gives you a ready-made onboarding prompt that explains the link to an assistant.
- Paste it where you work:
- at the start of a chat: "Here is my Memex: [link]. Check it before answering."
- in a Claude Project or a custom GPT's instructions, so every chat starts with it,
- in a coding agent's instructions, so it can look up earlier debugging sessions.
- Ask normally. "What did I conclude about pricing last month?" now has an answer.
What the assistant can and cannot do
By default a Memex link is read-only. The assistant can search and read; it cannot delete or change anything. You can optionally allow writes, which lets an agent:
- leave breadcrumbs, short notes recording which conversation helped with which question, so future assistants find the good material faster (see AI breadcrumbs);
- archive the current session, so a conversation held in one assistant becomes part of the memory every other assistant can use;
- attach files it produced to that archived conversation.
Access is metered and the link is revocable: if it leaks, revoke it and create a new one.
Example
Without a link, a new assistant asked "What should I work on next?" replies with generic questions about your goals.
With a link, the same assistant searches your archive, finds last week's planning session, and replies with something like: "Finish the migration API you scoped last week. You decided the token-gated link was the unlock, and the ingest pipeline is already done." Same model, same question, different starting point.
Good practices
- Treat the link like a password. Share it with assistants you trust and revoke it when you are done with a tool.
- Tell the assistant to search first. A line in your project instructions such as "check my Memex before answering questions about my work" makes the behavior consistent.
- Keep the archive current with the Memex Capture extension, so the link reflects this week, not last quarter.
- Prefer MCP where available. In clients that support it, the Memex MCP server gives the assistant proper tools instead of raw URLs.