Introducing Trilix: from scattered AI chats to one reasoning graph
Trilix turns linear AI chats into a branchable graph you can fork, run in parallel, and trace. Watch the demo, try the 60-second tutorial, and find us on Product Hunt.
If you use AI for real work, you probably have a dozen chat tabs open right now. One has the first draft. Another has the version where you tried a different model. A third is where you asked "what if we did the opposite?" and got the answer you actually needed, somewhere around message forty.
Chat is a good way to ask one question. It's a poor place to think, because thinking branches and a chat log only scrolls.
Trilix is a workspace built around that branching. Every prompt is a node on a canvas. You can fork any answer, run the branches side by side on different models, and see exactly which context produced which result.
Watch the demo
This video goes from a pile of chat tabs to a single reasoning graph:
Everything in the video is the real app. Nothing is a mock-up.
Four things you can do
Ask. Every question starts a graph. Pick a model for the node and run it. Answers come back as structured output: a summary, a confidence score, and detail you can collapse. Actions, risks, and next steps appear beside the answer, and each next step can become its own branch.
Branch. Drag from any answer to start a new line of thought. Each node can use a different model, so you can put GPT, Claude, and an open model on the same question and compare. Trilix logs the model, latency, tokens, and cost for every run.
Explore. Run several branches at once and they stream in parallel, side by side. You don't have to finish one direction before you try another.
Trace. Each branch carries only its own lineage as context: the nodes above it, and nothing from its siblings. Two branches exploring opposite ideas don't contaminate each other. One keystroke saves the whole graph, so you can come back to the exact path that produced a result.
Learn it in 60 seconds
The tutorial covers the four moves (ask, branch, explore, trace) in one minute:
Why a graph?
A chat log hides the alternatives you tried and dropped, and those are often the most useful part. In a graph they stay where you left them. You can see which prompt, which model, and which earlier answers led to a conclusion, and you can rerun that path later.
That record matters more as AI moves into work that has to be explained: research, consulting, product decisions, anything a colleague or reviewer will ask "how did you get here?" about. Trilix keeps the answer to that question with the work itself.
Try it
- Open the workspace.
- Sign in with GitHub or Google to get a few free model runs a day, or connect your own provider. Connecting a model to Trilix walks through every option, including free ones.
- Type a question, run it, and drag from the answer to branch.
Trilix is in free beta and we'd like to hear what works and what doesn't.
- Product Hunt: we're on Product Hunt. An upvote or a comment there helps other people find us.
- YouTube: more walkthroughs are coming on the Cognitive Interface Lab channel.
- Feedback: use the support page, or open an issue on GitHub.