Konshus University
Field notes on memory, portability, and you.
Konshus University is where we work out loud. Every piece here is an honest take on a real question we get asked — about AI memory, about what each model actually remembers, about what happens when providers change their roadmap, and about the version of you that gets to outlast any single tool. No fluff, no SEO sludge. Just the thinking behind the vault.
Pillar reports
Start here.
Report · Refreshed quarterly
The AI Memory Report 2026
Definitive annual audit of ChatGPT, Claude, Gemini, Perplexity, Copilot & Grok memory — retention, model-swap churn, portability, and the playbook.
Read the report · 22 min →
Standard · Open schema
The AI Memory Portability Standard
Our proposed open JSON schema for portable AI memory: atoms, artifacts, and provenance — with an inline reference implementation.
Read the report · 18 min →
Playbook · 10 chapters
The Complete Guide to Owning Your AI Consciousness
Every failure mode, every provider's export path, and the full ownership playbook — the one guide to send anyone starting to take this seriously.
Read the report · 35 min →
Flagship report · Jul 2026
The State of AI Memory 2026 (H2)
Mid-year scorecard across ChatGPT, Claude, Gemini, Copilot, Perplexity & Grok — what changed, what didn't, and where the ownership fight stands.
Read the report · 22 min →
Encyclopedia · Living reference
The AI Memory Troubleshooting Encyclopedia
The one page to send anyone whose AI memory just broke. Diagnostic tree, index by symptom and by vendor, and the structural prevention.
Read the report · 25 min →The catalog
Everything else, by category.
ChatGPT memory not working? 8 fixes that actually work
Deep 8-step diagnostic to restore ChatGPT memory — beyond the settings-check tips. HowTo schema for AEO.
Read →Why does ChatGPT forget me?
The structural reasons memory is fragile — the cap, the retrieval lottery, the Temporary-Chat trap.
Read →AI memory is wrong about you — the 4-step fix
Corrupted inferences, silent overwrites, confabulated recall — and the exact fix that makes corrections stick.
Read →AI personality changed after an update?
Why every major model release shifts your AI's vibe — and the voice-contract + few-shot fix that gets 80% of it back.
Read →Character.AI memory: how it works and how to make it stick
The three memory layers in Character.AI, why bots forget your name at message 400, and where to put facts so they survive.
Read →Perplexity memory: the complete guide
Threads, Spaces, and the Memory layer — how the three fit together and how to configure them for research that survives sessions.
Read →Gemini memory: the complete guide
Three layers — saved info, Workspace context, per-chat retrieval — with configuration for continuity without over-sharing.
Read →Copilot vs ChatGPT memory: the honest comparison
Consumer Copilot vs M365 Copilot vs ChatGPT — three memory models compared, and which to use when.
Read →Grok memory: how it works and how it compares
Cross-chat memory + X integration — the tradeoffs that make Grok's model structurally different from ChatGPT's.
Read →AI memory vs context window, explained
The most common misconception in AI, in plain English — with a decision tree for which failure mode you're actually seeing.
Read →Memory vs custom instructions vs system prompt
The three control layers, exactly which to use when, and why 'my instructions aren't working' usually means they're in the wrong layer.
Read →How much does ChatGPT actually remember?
1,000 controlled prompts across memory ON/OFF; recall rate over 30, 60, 90 days.
Read →ChatGPT memory deleted: what to do next
Recovery playbook when OpenAI wipes your memory. Real cases, screenshots, the fastest way to get context back.
Read →Claude 4 vs Claude 3: what actually changed
Model-transition case study. Charts of what Claude 'forgets' between versions from our own re-query benchmarks.
Read →Exporting every AI tool in 2026
Step-by-step exports for 15+ AI products with a file-format reference table and consolidation playbook.
Read →What happens when OpenAI shuts down a model
Historical case studies, notice-period chart, and the migration playbook for when yours is next.
Read →The hidden cost of switching AI models
Hours to rebuild context, quality-drop chart, and break-even math for portable memory.
Read →AI journaling vs AI memory
Day One AI, Rosebud, Mem, Reflect vs persistent memory — feature matrix, when each wins.
Read →How to audit what AI knows about you
30-minute self-audit across ChatGPT, Claude, Gemini, Meta AI. Deletion links + keep-or-wipe rubric.
Read →Persona drift: why AI feels different over time
Original research: tone and preference drift across six months of the same controlled prompt.
Read →The complete AI backup guide
Every AI tool, every export path, every gotcha — the one page to send anyone who asks how to back up their AI.
Read →AI Personality Test
Paste a few replies from your AI. See its tone across 6 axes, its archetype, and how it compares to other people's AIs.
Read →Konshus vs ChatGPT Memory
What ChatGPT remembers, what it forgets, and why portability changes the math.
Read →Konshus vs Claude Projects
Project-scoped containers vs a life-scoped vault — when each one wins.
Read →Best AI Memory Systems in 2026
An honest survey of ChatGPT Memory, Claude Projects, Mem, Rewind, and Konshus.
Read →How to back up ChatGPT
Step-by-step: export your full ChatGPT history before the next memory reset.
Read →ChatGPT memory, in full
Everything ChatGPT Memory actually does, and where it falls short.
Read →When AI updates wipe your memory
Why model upgrades silently erase context — and how to survive them.
Read →The lost AI memory problem
Why every AI you've ever used has forgotten you, and what to do about it.
Read →AI personal assistants
The four shapes of AI assistance and where a portable persona fits.
Read →Second brain, in the AI era
Notes apps were the old second brain. Here's what the new one looks like.
Read →AI prompts guide
How to write prompts that actually use your context — not just your question.
Read →How much does ChatGPT actually remember? (field note)
Real numbers behind ChatGPT Memory — about 1,200–2,000 tokens, 8–12 facts. Less than you think.
Read →The hidden cost of switching AI models (field note)
Every model swap quietly resets what your AI knows about you. 7–13 hours of retraining per switch.
Read →What happens to your AI when you die?
Photos and email have legacy programs. AI conversations don't. Here's what to do about it.
Read →How to back up your Limitless pendant
Export your Limitless transcripts before the cloud, the company, or your subscription decides for you.
Read →Readwise highlights → AI persona
Years of highlights are the closest record of your taste. Here's how to actually use them with any AI.
Read →Notion as AI memory
Export your workspace as Markdown and turn years of pages into a persona any AI can use.
Read →Obsidian vault → portable persona
Your vault is already portable. Here's how to make the AI layer portable too.
Read →Day One journals + AI
A decade of journal entries can teach a model the shape of you. Here's how to do it carefully.
Read →The ambient capture stack
Limitless vs Plaud vs Granola — what each is good at, and the lock-in problem they all share.
Read →How to export your ChatGPT memory
Walkthrough of OpenAI's data export — what's actually in it, what's missing, and what to do with the file.
Read →Claude Projects & model deprecation
What survives when Anthropic retires a model — and how to back up the parts that don't.
Read →When your memory app dies: Rewind
Rewind didn't shut down — it pivoted. For a memory app, that's almost the same thing. The lesson for the rest of us.
Read →Mem vs Notion vs Obsidian for AI memory
An honest comparison: AI UX, structure, portability, and lock-in. Which one (plus what) to pick.
Read →A running list of AI models that got wiped
Evergreen ledger of model deprecations, sunsets, and quiet swaps across OpenAI, Anthropic, and Google.
Read →Can you transfer ChatGPT memory to Claude?
Short answer: not directly. The manual path that works, what gets lost, and the cleaner version.
Read →AI memory comparison 2026 — 8 systems, 12 axes
The benchmark table: ChatGPT, Claude, Gemini, Mem, Rewind, Limitless, Notion AI, Konshus — scored on recall, portability, and control.
Read →The state of AI memory in 2026
15 events that reshaped what AI remembers about you — 2023 to today. The timeline nobody else has compiled.
Read →AI memory glossary — 40+ terms
Every term you need to talk about AI memory clearly: context window, RAG, atoms, distillation, persona export, and more.
Read →Best AI with long-term memory
Scored listicle: which tools actually remember, which fake it, and which let you take it with you.
Read →AI with persistent memory — native vs dedicated
When provider memory is enough, and when you need a dedicated memory layer underneath every AI you use.
Read →Character.AI memory — what it keeps, what it loses
How c.ai context windows work, why characters drift, and how to back up the personas you've built.
Read →ChatGPT memory not working? Fix it
The settings, the caps, and the silent resets — diagnose why ChatGPT keeps forgetting you.
Read →Perplexity memory — Threads, Spaces, profile
How Perplexity's three memory surfaces actually work together, and where the gaps are.
Read →Replika's 2023 memory loss
When a personality shift erased relationships overnight — and what it taught us about owning the memory of you.
Read →MCP and the end of provider-locked AI memory
What the Model Context Protocol is, who actually uses it, and why it's the first real portability primitive — with 2026 adoption numbers.
Read →Why every AI forgets you
Context window vs memory vs RAG vs fine-tuning vs system prompt. The five things people call AI memory, and why none of them are.
Read →The 2026 AI memory migration playbook
Export from ChatGPT, Claude, Gemini. What survives, what doesn't, and the portable-layer pattern that outlasts the next model retirement.
Read →Persona drift — when the model updates into a different friend
Why GPT-5 → 5.5 felt like a new therapist reading your old intake form. The mechanism, the documented 2025–2026 events, and what to do.
Read →