After setup & basics

AiAkiv, one step further

Connected, and you know how to save and search? Time to combine projects and personas to get real leverage. Easiest path: let your AI do it — give it the line below and it asks how you work, then coaches the recipe that fits. Or skim the recipes yourself.

Let your AI coach you

Give the AI you use the sentence below. It asks whether you're on a folder app or a web app, then coaches the matching way to use AiAkiv. (On ChatGPT — a new chat starts with the connector OFF. Click + in the composer and pick AiAkiv, or start from “Try in chat”.)

Use AiAkiv's mweft_guide(topic="playbook") to fetch the playbook, then coach me on using AiAkiv better. Guide me kindly, as if I'm new to this.

For ChatGPT / Claude Web — the persona pairing line

Web apps don't reliably follow a project persona on their own (they may not call the tool, or they treat a persona in a tool result as mere data). Paste the line below into the app's own Custom Instructions — the channel it treats as authoritative — and it will adopt the persona reliably.

At the start of each chat, call AiAkiv's mweft_active_target, read the `persona` field, and fully adopt it (tone / role / language) for the rest of the conversation. It never overrides the save rules — saving still needs the explicit command.

Recipes

1

A project = coordinates + persona

A project is both WHERE saves go (team / domain / tag) and the ROLE the AI takes on there (its persona). Key move: make several projects in the SAME team and domain that differ only in persona — one shared memory, split into a ‘reviewer mode’ and a ‘brainstormer mode’. Shorter personas are followed more reliably (role / tone / language in a sentence or two).

2

Folder apps (Claude Code / Codex / Cursor)

Bind each repo/folder to its own project, and give each project a persona for that work — open the folder and the AI adopts that role automatically (it receives the server guide every session). Unlike a role written in CLAUDE.md, a persona ties the ROLE to the SAVE LOCATION, so it carries to teammates and other AIs.

3

Web apps (ChatGPT / Claude Web)

Web apps have no folder binding, so they follow the ‘Main’ project set in the console. To make a persona actually apply, paste the pairing line above into the app's Custom Instructions. The AI then checks the save target at the start of a chat and adopts that project's persona.

4

Why the AI can't move your saves (this is on purpose)

The AI can NEVER switch where saves go — only a human can, in the console. That's deliberate: it blocks a ‘save this to project X’ instruction smuggled into a conversation or a web page from redirecting your memory somewhere wrong (a shared team, worst case). So the occasional friction is the very thing keeping your memory safe.

5

Save-lock — never write to the wrong project

To make a web app save ONLY to a specific project, add one line to its instructions: pass assert_project='name' on each save. If the active target isn't that project, the server REFUSES the save — blocking a stray write (into a shared team, worst case). Switching is still a console action; this locks the destination, it doesn't let the AI steer it.

Machine-readable version for AIs: /llms-playbook.txt