Feature Request: Versioned Project Source Sets
Summary
I would like to propose a relatively small extension to ChatGPT Projects:
Allow users to save a reviewed set of past ChatGPT messages as a versioned Project Source Set, preserve links to each original message, and review only the diff when updating the set.
This is not intended to replace Chat Search, Memory, Projects, or existing ideas around context checkpoints and conversation export.
The goal is narrower:
Search history → select relevant records → review them → save the exact set as Version 1 → later review only new or changed candidates → create Version 2.
The Problem
After using ChatGPT for months or years, useful information becomes distributed across many conversations.
For example, a long-running project may contain relevant information in:
-
old chats,
-
archived chats,
-
different Projects,
-
uploaded documents,
-
images,
-
corrections made months later,
-
decisions that replaced earlier decisions.
Chat Search can help find this material.
Projects can provide persistent context.
Memory can reuse relevant information.
However, I do not see a simple way to say:
“I have reviewed these exact historical messages. Save this specific collection as the source set for this Project, and do not silently change it.”
That distinction becomes important for long-term work.
Proposed Workflow
1. Search past ChatGPT history
The user searches for relevant historical records.
Example:
Find messages from 2024–2026 related to a specific long-term project.
2. Select relevant messages
The user selects messages from multiple conversations.
For example:
118 messages selected
3. Save them as a Project Source Set
Example:
Project History — Version 1
Each item keeps a reference to its original source message.
Possible metadata:
-
conversation ID,
-
message ID,
-
timestamp,
-
user / assistant role,
-
original message,
-
attachments,
-
images,
-
link to the original chat.
Version 1 is now fixed.
New chats should not silently modify it.
Updating the Source Set
Six months later, the user selects:
Update Source Set
ChatGPT searches for potentially relevant records added since the previous version.
For example:
12 new candidate messages found
The user reviews only the difference:
-
9 — Add
-
3 — Ignore
ChatGPT then creates:
Project History — Version 2
Version 1 remains available.
This makes it possible to answer questions such as:
-
What source material did I approve six months ago?
-
What has been added since then?
-
Which source set was used for this report?
-
Why did a later analysis use different evidence?
Why Versioning Matters
A dynamic search result is useful for discovery, but it is not the same as a reviewed historical record set.
Search results can change as:
-
new conversations are created,
-
retrieval improves,
-
classifications change,
-
search behavior changes.
For some workflows, this is desirable.
For others, the user needs a stable reference point.
The proposed Source Set would therefore behave more like:
a reviewed snapshot of source material
than a Smart Folder or dynamic search result.
Changes would create a new version instead of silently modifying the old one.
Source Traceability
Every item in a Source Set should remain linked to its original ChatGPT message.
This is especially important when ChatGPT later summarizes or interprets those records.
A user should always be able to move from:
AI-generated summary
back to:
original source message.
The Source Set should preserve the source rather than replacing it with a summary.
Difference from Chat Search
Chat Search answers:
Where did we discuss this?
A Versioned Project Source Set answers:
Which exact historical records have I chosen to use as the source material for this Project?
Search provides discovery.
Source Sets provide selection, preservation, and reuse.
Difference from Memory
Memory helps ChatGPT decide what previous information may be useful in a current conversation.
A Source Set would instead give the user explicit control over:
-
what is included,
-
what is excluded,
-
which version is being used,
-
where each record originally came from.
This is not intended to replace Memory.
It serves a different purpose: source traceability and controlled context selection.
Difference from Projects
Projects already provide the natural location for this feature.
A Project can contain many conversations and files, including:
-
outdated assumptions,
-
rejected ideas,
-
temporary discussions,
-
contradictory information.
A Versioned Source Set would allow the user to define a smaller, reviewed subset of those materials for a particular purpose.
For example, one Project could contain:
-
Research Sources V3 -
Approved Decisions V2 -
Customer Requirements V5
without having to duplicate the original chats.
Difference from Context Checkpoints
There are already strong proposals around auditable context checkpoints containing facts, corrections, decisions, rejected hypotheses, and source links.
I do not intend to claim those concepts as new.
The distinction I am proposing is mainly scope.
A Context Checkpoint can represent:
the verified working state of a conversation.
A Versioned Project Source Set would represent:
a reviewed collection of source records gathered across many conversations.
The two could complement each other.
Example Use Case
Imagine a parent who has used ChatGPT for several years to discuss a child’s development.
Relevant records may be scattered across conversations about:
-
school,
-
therapy,
-
home life,
-
learning,
-
photographs,
-
direct quotes,
-
parent observations,
-
previous AI analyses.
The user searches the history and selects 118 relevant records.
They save:
Child Development History — V1
Six months later, ChatGPT finds 12 new candidate records.
The user accepts 9 and rejects 3.
ChatGPT creates:
Child Development History — V2
The user could then ask:
“Create an annual summary using Child Development History V2.”
or:
“Compare V1 and V2 and show what was added.”
The same workflow could apply to research, software projects, procurement, learning records, business decisions, or other long-running work.
Important Limitation
This feature would not guarantee reproduction of the same AI output.
Even when the same Source Set is used, results can change because of:
-
model updates,
-
system behavior,
-
prompt changes,
-
tools,
-
retrieval behavior.
What the Source Set preserves is:
the exact input evidence set selected by the user
—not the entire model execution environment.
So the benefit is better described as input traceability, not complete output reproducibility.
Human Review Does Not Mean Truth
A reviewed Source Set should also not be described as automatically “verified truth.”
User approval means:
This is the source set the user chose to use at that time.
It does not prove that every statement inside it is objectively correct.
For that reason, I prefer Reviewed Source Set rather than Verified Source Set.
Deleted Sources
If an original conversation or file is later deleted, user deletion rights should take precedence.
The older Source Set could retain a marker such as:
Source unavailable
rather than silently pretending the source still exists.
The version definition can remain historically visible without preserving content the user has chosen to delete.
Minimum Viable Version
A first implementation could be very small:
-
Select multiple messages from Chat Search.
-
Add them to a Project Source Set.
-
Save the set as V1.
-
Preserve links to the original messages.
-
Re-run the search later.
-
Show new candidates as a diff.
-
Let the user Add / Ignore them.
-
Save V2.
-
Allow any version to be used as explicit Project context.
Advanced semantic classification would not be required initially.
Relationship to Existing Ideas
I am not claiming novelty for:
-
chat-history search,
-
semantic retrieval,
-
selective export,
-
context checkpoints,
-
source citations,
-
Smart Folders,
-
version control,
-
editable conversation timelines.
The narrower product idea is:
Connect Chat Search and Projects with a reviewed, versioned source-set workflow.
In short:
Search once → review once → reuse many times → review only the changes.
One-Sentence Proposal
Let users save a reviewed set of past ChatGPT messages as a versioned Project Source Set, preserve links to each original message, and review only the diff when updating the set.