# AiAkiv

> A long-term memory that several AI clients - and teammates - read and write
> over MCP (Model Context Protocol), so the context built with one AI carries to
> the next instead of vanishing when a chat ends.

Connect first, then read. The setup file below is a SCRIPT to run with the user,
not a page to paste.

LANGUAGE: these docs are authored in Korean and served in the reader's language.
Every URL below negotiates on Accept-Language. To pin a language, put its code
after /docs - `/docs/ko/saving`, `/docs/en/saving`. Answer the user in whatever
language THEY write in, regardless of which language you fetched.

## Setup - do this first

- [Interactive setup](https://www.aiakiv.com/llms-setup.txt): run this with the
  user to connect their AI client. Ask, branch, configure, verify - in that order.

## What each connection has

There is one server, https://mcp.aiakiv.com/mcp. The tool list depends on how the
client attached.

- Every connection has: saving and searching memory, reading original text,
  reading linked teams, graph queries, and creating cards (`create_aiakiv_card`).
- App features (asking the AI to do app work such as console management, through
  `run_aiakiv_app_action`) are on folder connections, API key connections, plugin
  installs in Claude Code and Cursor, and connections attached with the address
  `https://mcp.aiakiv.com/mcp?apps=all`. That address is the same endpoint with
  app features turned on. It is the address to enter wherever an address is typed
  in and the user signs in: the Claude connector (web, desktop, mobile), Grok, a
  Codex connection set up with the install prompt
  (https://www.aiakiv.com/docs/codex-install-prompt), and similar clients.
- The official ChatGPT plugin does not support app features. There are two ways
  to use them: the Codex connection above (OAuth sign-in, account-wide, follows
  Main), or a project-pinned API key in a folder's `.codex/config.toml`.
- App features are also absent on a sign-in connection attached with the address
  without `?apps=all`, such as a Claude account connector added that way earlier.
  Such a connection stays as it is until the user changes its address to the
  `?apps=all` address, or removes it and adds it again. A tool list without
  `run_aiakiv_app_action` means that connection lacks them, and console work is
  then done by the user in the web console.
- When one conversation has two connections with `run_aiakiv_app_action` (an
  account connector on the `?apps=all` address next to a folder connection), the
  folder or API key connection is the one to use for it.
- The console app covers teams and invitations, team links, audit and save target
  history, projects and personas, Main (the save target), public targets, the
  user's cards, and a read of the AI permissions. All of it can also be done by
  the user by hand in the web console (the dashboard, https://app.aiakiv.com);
  asking the AI is a convenience.

More: [The app action tool](https://www.aiakiv.com/docs/app-actions) and
[Connecting ChatGPT](https://www.aiakiv.com/docs/chatgpt).

## Start here — what it is

- [What AiAkiv is](https://www.aiakiv.com/docs/what-is-aiakiv): AiAkiv is an outside service that remembers relationships, not a file locker, but a long-term memory for AI that records what happened and why.
- [Core concepts: Event · Entity · Tag](https://www.aiakiv.com/docs/core-concepts): Saving creates three things. Knowing who makes each one tells you what to care about: why entities keep search alive, and why tags are worth attaching.
- [AiAkiv is not a document store](https://www.aiakiv.com/docs/not-a-docstore): What gets stored is shaped for an AI to read. Keep the documents people read somewhere else, and leave in AiAkiv the process that produced them.

## Using it

- [Five-minute quickstart](https://www.aiakiv.com/docs/quickstart): Connect, save once, search once. Three steps is enough to see AiAkiv working.
- [Saving](https://www.aiakiv.com/docs/saving): It saves only when you command it with `ak`. The conventions that make memories worth having, and how to correct or retire one afterwards.
- [Searching](https://www.aiakiv.com/docs/searching): It finds by meaning and connection, not keyword matching. `@` attribution, causal and time questions, and how to read a result.
- [Chained saves (threads)](https://www.aiakiv.com/docs/threads): Chain, don't trim. Split long text and ongoing records with `prev_event_id` so the flow survives.
- [Asking by relationship: ak graph](https://www.aiakiv.com/docs/graph-query): Graph queries with `ak graph`: fix a name or tag scope, then ask. How to ask, example requests and the queries they become, the grammar, how to read the response, and the limits.

## Boundaries

- [Teams and projects: the team owns the memory, the project is the way in](https://www.aiakiv.com/docs/teams-and-projects): Who owns the memory (the team) and how you get into it (the project). The save target (Main) and the guard on switching it.
- [Creating and managing teams (including sub-parts)](https://www.aiakiv.com/docs/teams): Invitations and roles, the sub-part tree, reading flows down but writing does not, the ‘Apply’ gate, and retire vs delete.
- [Team sharing](https://www.aiakiv.com/docs/sharing): AiAkiv's real strength shows when several people build the memory together. Team isolation and `@` attribution keep it from blurring.
- [What you may store (security and limits)](https://www.aiakiv.com/docs/what-to-store): It's an outside service and what you store stays. The do-not-store list, where responsibility sits, and the domains an IT admin blocks.

## Connecting your AI

- [Authentication: OAuth and API keys](https://www.aiakiv.com/docs/auth): OAuth has the system authenticate for you; an API key binds authentication to the key itself. Which client uses which, and the minimum key hygiene.
- [Folder binding: pinning a save target per working folder](https://www.aiakiv.com/docs/folder-binding): Pin a project to a folder and the switching accident disappears structurally. Per-client setup and how to read the `binding` value.
- [Using project binding in Claude Desktop](https://www.aiakiv.com/docs/desktop-binding): `</> Code` is Claude Code. But it never shows the approval prompt: without approving in advance only the account connector is left and saves go to Main. Even when approved the account connector attaches alongside; leave it on and check where each save went.
- [When the account connector and a folder connection are both attached](https://www.aiakiv.com/docs/account-connector): Claude Code and Desktop's `</> Code` attach your claude.ai account connector separately from the folder connection. Leave the account connector on: turning it off removes AiAkiv from the web and the mobile app too. The AI saves through the folder connection first, but that isn't enforced, so check where each save went. If saves must go to the folder only, turn Main off for a while in the AiAkiv console.
- [Connecting ChatGPT and ChatGPT classic](https://www.aiakiv.com/docs/chatgpt): Both apps install AiAkiv from the plugin directory. To pin a project per folder, put a folder connection in the working folder, either the sign-in (OAuth) way or the API key way. App features are not in the official ChatGPT plugin, because parts of them don't fit OpenAI's plugin review policy. To use them, connect Codex separately with the install prompt (app features and seven skills, saving to Main) or use a folder connection.
- [The Codex install prompt: AiAkiv connection and skills in one go](https://www.aiakiv.com/docs/codex-install-prompt): Paste it into a Codex chat and Codex registers the AiAkiv connection, opens the sign-in window, installs the seven skills and writes an install record on its own. No developer mode needed. Before pasting, remove the existing AiAkiv Memory plugin, or, if you keep it, rename this connection to AiAkivCodex.
- [The Codex verification prompt: checking the install](https://www.aiakiv.com/docs/codex-verify-prompt): Paste it into a new session after installing with the Codex install prompt. It checks the connection, the two app tools (create_aiakiv_card, run_aiakiv_app_action), the save location (binding main) and the seven skills one by one, and appends the results to the install record.
- [The AiAkiv CLI: pin a project to a folder with one command](https://www.aiakiv.com/docs/aiakiv-cli): Run npx aiakiv bind once to pin an AiAkiv project to your working folder. It writes the same config files and skills as the console's folder ZIP straight into the folder, and for Codex and Gemini CLI it also fetches the key after you press a confirm button in the browser. Check the result with npx aiakiv doctor. The account-wide install is done with npx aiakiv setup, for Gemini CLI and Antigravity only.

## Reaching further

- [Links: reading another team's memory](https://www.aiakiv.com/docs/links): Two memories are joined by entity aliases. The link is one-directional, and when memory content changes the aliases have to be updated too.
- [Public memory: memories anyone can read](https://www.aiakiv.com/docs/public-memory): Anyone attaches and reads without an invitation, but nobody writes, and only AiAkiv can create one for now. The two ways to attach, and the public attach that ties one into your own team.
- [Economic-Report: an economic data space everyone can read](https://www.aiakiv.com/docs/economic-report): A read-only public memory of public economic data and institutional releases. Eighteen source projects, how to use it, and the data caveats before citing.
- [Cards: pulling a memory out as one shareable page](https://www.aiakiv.com/docs/cards): A one-page public page your AI writes from your memories. Public the moment it's made; managed from the console or by asking your AI.

## With AI, and when stuck

- [For AIs connected to AiAkiv](https://www.aiakiv.com/docs/for-ai-agents): Written for an AI, not a person. Three rules, and a map of every tool the engine exposes.
- [Working with several AIs (and still understanding it)](https://www.aiakiv.com/docs/working-with-ais): Share judgment, not just output. Splitting implementer and reviewer, handing off through ak, and staying a node inside the graph rather than outside it.
- [Console menu guide](https://www.aiakiv.com/docs/console): Saving and searching happen in your AI; the console is where you build and maintain the ways in. What each of the six menu groups does.
- [Asking your AI to do console work](https://www.aiakiv.com/docs/console-actions): Teams, links, audit, history, projects, the save target and cards from the conversation. You decide what is turned on in AI permissions; anything irreversible still belongs to a human.
- [The app action tool: run_aiakiv_app_action and which connections have it](https://www.aiakiv.com/docs/app-actions): The tool your AI uses to ask an AiAkiv app (such as the console app) to do something. Folder connections, API key connections, plugin installs and connections attached with the address `https://mcp.aiakiv.com/mcp?apps=all` have it; the official ChatGPT plugin and sign-in connections attached with an address that lacks `?apps=all` do not. What you can do without it, and which connections have it. In ChatGPT you get it by connecting Codex with the install prompt or by connecting a folder with an API key.
- [Common problems (FAQ)](https://www.aiakiv.com/docs/faq): When a memory you expected isn't there, when saving doesn't happen, when search comes back wrong. What to check, in order.
- [Saving tokens, Claude only: work in chunks, save, then compact](https://www.aiakiv.com/docs/saving-tokens): Claude only. Six habits (work in chunks, keep one model, return within 1 hour, ak save, compact, rest) that keep the earlier conversation in memory and add only what changed: fewer tokens, context kept. Does not apply to ChatGPT.

## Optional

- [Full text of every doc](https://www.aiakiv.com/llms-full.txt): everything above concatenated, for one fetch instead of many.
- [Pricing](https://www.aiakiv.com/pricing): every account has one personal memory (10 new memories a day, no total limit); plans differ in the number of team memories, their daily new saves and their total. Reading, searching, editing and exporting are never blocked. Every feature is free during the preview.
- [Tutorial](https://www.aiakiv.com/llms-tour.txt): first-run walkthrough.
- [Advanced workflows](https://www.aiakiv.com/llms-playbook.txt): projects and personas.
- [Console actions](https://www.aiakiv.com/llms-console.txt): manage teams, links, audit and cards from the conversation.
- [Graph query](https://www.aiakiv.com/llms-graph-query.txt): AiAkiv graph queries: the ak graph request, the query subset, and reading the result.
- [Saving tokens](https://www.aiakiv.com/llms-saving-tokens.txt): AiAkiv and saving tokens with Claude: working in chunks, saving each chunk, then compacting.