Selective "User-Only" Consolidation (The Clean-Slate Branch)

I’m proposing a power-user feature called "Consolidated Input Branching" (or "User-Only Context").

Currently, T3 Chat handles linear branching well, but it forces us to carry over the entire history (User + AI). I propose a "Context Filter" that allows us to branch using only the User's input history, stripping out previous AI responses.

The Strategic Alignment (Why this matters now):
Theo has mentioned (on X) that ~99% of users don't exhaust their 1500 message limit. This feature specifically targets that efficiency:

1. Drastic Token Reduction: Standard branching re-sends everything (including verbose AI replies) to the API. By stripping the AI responses and sending only consolidated User prompts, we drastically cut input token costs for T3.

2. BYOK Savings: For power users pushing limits or using BYOK, this prevents burning money on processing "dead" context just to switch models.

The Workflow Pain Point:
When iterating on complex logic, the AI often gets "polluted" by its own previous wrong answers. To fix this, I currently have to:

1. Scroll back and copy 5 different user prompts.

2. Paste them into Notepad.

3. Start a fresh chat to get a "clean run" with a new model.

The Proposed Solution:
Add a "Consolidate User Inputs" option in the Branch menu.

Dev Logic:

1. Filter:messages.filter(msg => msg.role === 'user')

2. Merge: Concatenate user inputs (sanitized).

3. Inject: Start a new thread with the Model Selector, using the clean, merged text as the system/first prompt.

The "Win-Win-Win":

• For Users: It eliminates the "Notepad loop," making T3 Chat the definitive "Big-Brain" brainstorming platform for rapid cross-model testing.

• For T3: It reduces "Wasteful Tokens." You aren't paying for the API to re-read its own hallucinations.

• For The Ecosystem: It solidifies T3 Chat’s position as a professional workbench, keeping users on the platform rather than exporting to local notes.