Hold competing hypotheses apart.
The failure mode of research chat is contamination: by the time you have argued three theories in one thread, the model's answers are a smoothie of all three. Branches keep each argument clean enough to judge.
One anchor, three theories, zero cross-contamination.
Researching competing hypotheses
Set up the shared evidence once, then branch once per hypothesis off that anchor. Each branch argues its case against the same facts without hearing the others. When the arguments are long, summarize each branch before judging, and accept the honest outcome when it arrives: sometimes the evidence cannot distinguish them yet.
- A shared anchor message holds the common evidence
- One branch per hypothesis; arguments never bleed
- Summarize each branch, then compare the summaries
Literature reviews without tab chaos
Give each source its own branch off the review's trunk: summary, method critique, and relevance in one place per paper. Project files hold the PDFs, project memory holds what you decided about each, and the Merge Board turns the field of readings into the related-work section.
- One branch per source, attached to the review trunk
- PDFs and notes live in the project, briefed to the model
- Merge Board synthesizes the branches into the review
What this workflow leans on.
- Branching: one hypothesis per path, all free
- Explore with the Judge: run a question across models, scored
- Projects with files: the corpus travels with every conversation
- Scoped memory: what you concluded per project, inspectable
- Merge Board: Summary and Decision outputs with reasons
- Export: Markdown out when the review ships
GPTree is built to support judgment, not replace it. Field guide · Comparing branches
Never re-explain yourself.
Start free in under a minute. No card required. Use built-in credits or bring your own keys.