Run your own AI · Switch models, keep your context

The model can change. Your context should stay.

thecontinuitylayer.com/open/learn/keep-your-context

Checked 6 November 2026. US dollars. No affiliate links.

Running your own AI means you will switch models often: a new Qwen this month, a new Gemma next. Every switch starts from zero unless your context lives outside the model. That idea is what this site is named for: a continuity layer, the part that stays when the model changes.

By the end

One file with who you are, how you work and what you are working on, that you paste into or attach to any model, local or hosted.

What to keep

Who you are and what you do, in three or four lines.
How you want answers: length, tone, format, what to skip.
What you are working on now, and the decisions already made.
Facts the model keeps getting wrong about you.

Set it up

1
If you use ChatGPT, Claude or Gemini, get out what it already knows about you. Ask it to list what it remembers, or use See what your AI remembers to turn its answer into a file.

Write the file in plain text, named something like me.md. A starting shape:

# About me
Sam. I run a small bakery and write its newsletter.

# How I want answers
Short. Steps as a checklist. No preamble.

# Working on now
- Moving my notes to a local model
- Decided: Gemma 4 12B for daily chat

3
Give it to your local model as standing instructions. In LM Studio, paste it into the system prompt. In Open WebUI, put it in a model's system prompt or add the facts to its memory. In Ollama, save it into a custom model with a Modelfile:
```
FROM gemma4:12b
SYSTEM """
(paste the contents of me.md here)
"""
```
then run ollama create me -f Modelfile and ollama run me.
4
When you switch models, change the first line (FROM ...) and run the create command again. Your context comes with you.
5
Update the file when something changes. It is yours: any model, any app, any company.

Go further

The full method, for a record that any model reads at the start and updates at the end, is in Build your own continuity layer. Which hosted assistants let you take your memory with you is graded on the Scorecard, and the memory tools you can self-host are on the memory-systems shelf.

← Model sizes and quantization Next: System prompts and custom models →