GeniffyDocs
Changelog Log In Get a key

What is Geniffy

Geniffy is the memory infrastructure for AI applications and agents. You, your agents and your apps send it what happens: notes, conversations, whole work sessions with their tool steps, files and web pages. It reads them the way a person would, and keeps where each project stands, what was promised, what happened and why, and what was learned, with the line each part came from. Any model, agent or AI app reads it back in one call, in milliseconds.

Three ways to use it

  • In your app. Give each of your users a brain of their own, from Python, TypeScript or any language over HTTPS. Start with the quickstart.
  • In the AI tools you use. Claude, ChatGPT, Cursor, VS Code and Codex read and add to one memory through the MCP server, and the Claude Code plugin opens every session with the briefing and saves it whole.
  • Through your coding agent. Paste one prompt into Claude Code, Cursor or Codex inside your project, and it wires Geniffy into your code.

What it keeps

  • Where things stand. Each project's now, next and still open, and what was decided and why.
  • Promises. What someone said they would do, and by when.
  • What happened. Whole conversations and agent sessions read as stories: what was tried, what failed, what worked.
  • Lessons. What worked and what didn't, so the next session doesn't repeat a mistake.
  • Facts. Each one with the exact line it came from, who said it and when.

Every night it settles what disagrees (each change with its reason, and an undo), merges what it learned, and tests itself. The briefing hands all of it to your model in one call, read rather than generated, so it comes back at once.

What makes it different

  • The state of the work, not a pile of facts. Asked where you left off, it answers with where things stand, what is due and what happened, not the nearest few facts.
  • Every part has a source. You can show where an answer came from, or check it.
  • It says when it doesn't know. When nothing stored supports an answer, ask() returns no answer and a sentence saying so, and context() tells your model not to guess.
  • Unlimited reads. Search and briefings are never counted, on any plan; a plan sets what it learns and the answers it writes. See Plans and usage.
  • One brain per user, nothing across. Spaces keep each of your users apart, and one call erases everything held for one of them.

How it works

1

In

Send a note, a conversation, a whole agent session with its tool calls, a file (PDF, Word, PowerPoint, Excel or text), or a web page. Each becomes a source, kept and searchable at once.

2

Learn

Geniffy reads the source: the facts, with the line each came from and the date it is about, then the story of what happened, where the project stands now, the promises made and the lessons learned.

3

Out

Before your model starts, briefing() hands it where things stand; before it answers, context() hands it what bears on the question, or ask() answers in words, with the sources the answer rests on.

4

Correct and forget

When something is wrong, say what is right; when someone asks to be forgotten, forget them. See Correct and forget.

The whole path, from what you add to what comes back, is in How the brain works. Geniffy stores memory in India. See Security and privacy. Your key reaches only your own memory and the spaces beneath it.

Last updated October 18, 2026