Context
This article is written from the perspective of a LLM acting as an Architect on the project.
*LLM Architect* = ChatGPT or similar
Med Manager – example web app project
Not a one chat
For a project like your app, I would not worry too much about keeping everything in a single chat forever.
What I can and cannot remember
I can remember some long-term information about you (for example, that you’re building this Med Manager project ), but I do not automatically retain the detailed state of every file, migration, controller, and Codex change across chats.
So the most important thing is not the chat itself—it’s having a project summary.
Recommended Workflow
Option A (Best): Maintain PROJECT_STATUS.md
Create a file in your repo:
PROJECT_STATUS.md
Update it whenever a major feature is completed.
Example:
# Med Manager
## Completed
- Authentication
- Doctors CRUD
- Conditions CRUD
- Medications CRUD
- Medication Feedback
- Prescription Uploads
## Database
conditions
- id
- user_id
- doctor_id
- severity
- status
medications
- id
- condition_id
- dosage
- frequency
- ...
## Current UI Status
Conditions:
- standardized
Doctors:
- needs redesign
Medications:
- partially standardized
## Next Priorities
1. Severity badges
2. Medication PDF
3. Dashboard
Then whenever you start a new chat:
Here’s my PROJECT_STATUS.md
and I’m immediately caught up.
When to Start a New Chat
I would start a new chat when:
1. Topic changes significantly
Current:
Laravel Med Manager
New:
French language practice
or
Taxes
or
Moving to …
Those should be separate chats.
2. The chat becomes cluttered
For example, after:
- 50 Laravel errors
- 20 Codex discussions
- 30 feature changes
the chat can become noisy.
At that point:
Start a new chat
Paste PROJECT_STATUS.md
Continue
This is often cleaner than scrolling through months of history.
A Good Pattern
I would keep:
Chat 1
Med Manager Development
Chat 2
Freelance / Business
Chat 3
French Learning
Chat 4
Taxes / Immigration
That makes things easier to find later.
For Your Specific Project
You are reaching the stage where the app is no longer a tiny CRUD exercise.
You now have:
- Users
- Doctors
- Conditions
- Medications
- Feedback
- Prescriptions
- PDFs
- Severity (being added)
- Tailwind UI standardization
- Codex-assisted development
So I would suggest:
Now
Ask Codex to generate:
PROJECT_STATUS.md
and keep it in the repository.
Later
Whenever a chat becomes inconveniently long:
- Start a new chat.
- Paste PROJECT_STATUS.md.
- Tell me:
- current goal
- current problem
- any recent Codex changes.
Then I can usually get up to speed in a few minutes, regardless of how many previous chats existed.
For a solo developer using both ChatGPT and Codex, a maintained PROJECT_STATUS.md is more valuable than trying to keep a single chat alive indefinitely.
