Playbook 2 of 5

Partition your AI use.

From one ChatGPT bag of everything to a properly scoped dequid setup. Import, map, clean.


The single-bag problem.

When you use one AI product for everything, it accumulates one bag of facts about you: your job title, your co-founders, your health habits, your writing style, your weekend hobbies. Every question pulls from the entire bag.

That's Context Bleed — the failure mode where one identity's memory shows up in another's chat. Your CEO email sounds like your personal journal because both live in the same flat context. Why this happens →

The fix isn't more memory — it's bounded memory. Personas are the container.

Four partition patterns.

Most people adopt one of these. The question isn't which looks right in theory — it's which matches how your work actually splits.

Most common

By audience

To CEOTo Direct ReportsTo Customers

Pick this when the same information needs to sound completely different depending on who reads it.

Stakeholder-led

By project

Project X LeadProject Y Contributor

Pick this when projects have non-overlapping stakeholders and you want each conversation to stay in its lane.

Life-led

By domain

Engineer-MeWriter-MeFounder-Me

Pick this when your life domains are clearly distinct and you want the AI to speak from that domain's expertise.

Composite

Hybrid

Project X Status for Leadership

Combines functional and audience axes. Use when a project has a fixed audience and you want both dimensions enforced.

Six months of ChatGPT, partitioned.

Here's how a typical mixed-use account maps to four dequid personas.

Before — one ChatGPT account
Engineering questions (debugging, architecture, code review)
Weekly CEO status updates
Personal health and lifestyle questions
Creative writing and blog posts
Customer pitch decks
One-on-one coaching with direct reports

One bag. Everything bleeds into everything.

Import your history.

1

Export from ChatGPT or Claude

In ChatGPT: Settings → Data controls → Export data. You'll receive a .zip with a conversations.json inside. In Claude: Settings → Privacy → Export data. You'll receive a .json file.

2

Open Settings → Data → Import Conversations

Drag or click to upload your .zip or .json file — dequid parses it client-side, so the raw file never leaves your browser.

3

Pick a destination folder

Choose which conversation folder to import into. If you have separate folders per persona (recommended), pick the right one. You can re-import into multiple folders if conversations span multiple personas.

4

Review extracted memories

After import, dequid automatically extracts memories from the imported conversations. Go to Memory → Memories to review what was extracted before assigning them to personas.

Post-import cleanup.

Three things to do once your history is in.

Review auto-extracted memories

Go to Memory → Memories. The extractor is thorough — delete anything that's wrong, duplicated, or that belongs to a different scope. Memories are global by default; they'll reach every persona until you exclude them.

Exclude by persona

Open each persona → Memory tab. Tick any memory phrase that shouldn't reach that persona. A health memory ticked off the CEO Update persona won't bleed through — ever.

Tag old conversations to folders

Move imported conversations to the right persona's chat folder. This scopes their content when the persona uses conversation history context, and keeps the sidebar organised.

Next playbook

SEEK recipes

Four-dial presets you can copy for code review, coaching, research, and more.

Read →