🧠✨ Connect Brain Dots
Messy thoughts in. Clear decisions out.
An open-source AI brain buddy that untangles your "but what if..." thoughts, researches the evidence,
and draws you a complete, interactive map of your options before telling you which one fits you best.
Watch the demo · See it in action · Run it yourself · Deploy for free
😵💫 The problem: deciding is hard
I kept struggling with decisions, big and small.
Take something as simple as lunch. "I'll cook rice and meat curry... but my husband doesn't like curry. So I'll fry the meat instead? But frying isn't healthy. Then what?" Every single thought drags three more along with it: worries, constraints, other people's feelings, what you read somewhere once. Now scale that up to "Should I do a PhD while AI is changing everything?" and your head turns into a ball of yarn.
Most AI chatbots just hand you one answer. You can't see why, what it ignored, or whether it would still hold if your priorities were a little different.
Connect Brain Dots does the opposite. It lays out every dot your brain is juggling, connects them with real evidence, and lets you explore the whole picture before it gives its verdict.
🎬 Watch the demo
▶️ Watch the demo on YouTube: from "Should I do a PhD while AI is doing everything?" to a full decision dashboard. (Prefer a file? Download Demo.mp4.)
✨ What it does
- 🧠 You dump the thought. Type it exactly like it sounds in your head. Messy is welcome.
- 🫧 Dot mirrors your side-thoughts. Your avatar appears with every worry, constraint and goal connected to its brain. Pop the wrong ones and add the ones Dot missed.
- 🔎 The agent does the legwork. It breaks the idea into weighted criteria, searches the web for studies, facts and real people's experiences, brainstorms genuinely different options, scores them, and fact-checks itself.
- 📊 You explore everything. Constraints, consequences, comparisons, what-if sliders, the research report and the full reasoning, all interactive.
- 🏆 Then the verdict. The best fit for you, why it beats the runner-up, the promising alternatives, and your first steps.
It works for tiny decisions (lunch, weekend plans) and big ones (career moves, thesis methods, research trade-offs). It remembers your profile and past choices, so it gets more personal over time.
📸 A tour of the app
1. Get started
| |
|---|
| |
| Landing page with animated brain dots | Sign up with just a username, email and password |
| |
| Build your avatar (hand-drawn Open Peeps) | Chat with Dot so it learns what matters to you |
| |
| Bring your own AI: Claude, OpenAI, Grok, Gemini, Ollama or Demo mode | Brain dump your messy thought |
2. Untangle the thought
| |
|---|
| |
| Your brain graph: every side-thought connected to you. Pop or add bubbles | Dot at work: research, scoring and fact-checking, live |
3. Explore the whole decision
The results page is a guided journey: the verdict waits at the end.
| |
|---|
| |
| The question and how much was analysed | Overview: every option's fit %, win % and constraints met |
| |
| Your constraints: which options meet or break each one | Consequences: "If you choose X..." tree plus a risk map |
| |
| Compare: scores, weights and where the points come from | What-if sliders: change your priorities, watch the ranking move |
| |
| Research report: filterable evidence and cited sources | Reasoning: the decision path and blueprint diagram |
| |
| Verdict: why #1 beats #2, alternatives and first steps | Works on mobile too |
🛡️ Why you can trust the answer
The AI never gets the final word on its own. A deterministic layer (src/lib/scoring.ts) sits between the model and the dashboard:
- Every claim needs a receipt. Scores must cite evidence or one of your considerations. Citations to things that don't exist are dropped, and uncited scores are flagged with a "?".
- No fake research. Evidence claiming to be research without a real source is downgraded to "weak".
- Math, not vibes. The ranking is calculated in code from weighted scores. The AI explains the ranking; it can't invent a different one.
- It checks itself. A separate critique pass audits the scores against the evidence and adjusts them, and you can see every change.
- Honest confidence. Confidence is capped by how much of the analysis is actually backed by sources.
- You can stress-test it. The what-if sliders and an 800-run simulation show whether the winner holds up when priorities shift.
npm run test:agent checks these guardrails against a scripted mock model, and CI runs it on every pull request.
This is the beginning of the README. Read the rest at the original. Read the full README ↗
