🧠✨ 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.
🚀 Run it yourself
Option A: Use your own free deployment
Follow DEPLOY.md to put your own copy online with Neon (free Postgres) and Vercel (free hosting) in about 20 minutes. Anyone can then sign up and use it with their own AI key.
Option B: Run it on your computer (about 5 minutes)
You need Node.js 20.9+ and Docker Desktop.
git clone https://github.com/RafflesiaKhan/connect-brain-dots.git
cd connect-brain-dots
npm install
docker compose up -d # starts a local Postgres database
cp .env.example .env.local
Open .env.local and set a secret (run openssl rand -base64 32 to make one):
AUTH_SECRET="paste-your-secret-here"
The database line already points at the Docker database. Then:
npm run db:push # creates the tables
npm run dev # http://localhost:3000
Open http://localhost:3000, click Create account, and choose Demo mode at the AI step to explore without a key.
Getting an AI key
Each user brings their own key and picks a model in Settings. Keys are encrypted (AES-256-GCM) before they're stored and never sent back to the browser.
Every provider offers 3 to 4 models, sorted from 🧠 Big brain (deepest reasoning, most tokens) to 🐣 Tiny (fewest tokens, cheapest), with context size and price shown. You can also type any other model id your provider supports.
For models without built-in search, add a free Tavily key in Settings to enable web research.
⚙️ Configuration
All settings live in .env.local (locally) or your host's environment variables. See .env.example for the full list.
| Variable | Required | What it does |
|---|
DATABASE_URL | ✅ | Postgres connection string (Docker locally, Neon online) |
AUTH_SECRET | ✅ | Signs login sessions and encrypts saved API keys. Never change it after launch |
AUTH_GOOGLE_ID / AUTH_GOOGLE_SECRET | | Adds a "Continue with Google" button |
AUTH_GITHUB_ID / AUTH_GITHUB_SECRET | | Adds a "Continue with GitHub" button |
TAVILY_API_KEY | | Shared web search for all users |
ANTHROPIC_API_KEY, OPENAI_API_KEY, ... | | Shared "house" AI keys. Leave empty on public sites, or every visitor spends your credits |
OLLAMA_BASE_URL | | Default Ollama server address |
OLLAMA_ALLOW_CUSTOM_URL | | Let users enter their own Ollama URL in production (off by default for security) |
🧩 How it's built
| Area | Choice |
|---|
| Framework | Next.js 16 (App Router), React 19, TypeScript |
| Styling and motion | Tailwind CSS 4, Motion |
| Graphs and charts | React Flow, Recharts |
| Avatars | Open Peeps by Pablo Stanley, rendered with DiceBear |
| AI | Vercel AI SDK: Claude, OpenAI, Grok, Gemini, Ollama |
| Research | Provider web search or Tavily |
| Database | Postgres via Drizzle ORM |
| Auth | Auth.js: username or email with password (scrypt hashed), optional Google/GitHub |
The agent pipeline
your thought ─▶ reflect ─▶ you edit the side-thoughts
│
┌──────────────────────────────┘
▼
1. untangle goal + weighted criteria, personalised to your profile
2. research web search: studies, facts, people's experiences
3. options 3-5 genuinely different paths, each criterion scored with citations
4. critique a skeptical reviewer pass adjusts weak scores
5. connect verdict, alternatives, first steps
│
▼
deterministic guardrails (scoring.ts) ─▶ interactive dashboard
Project map
src/
app/
page.tsx landing page
login/ sign up / sign in
onboarding/ avatar builder, profile chat, AI setup
home/ brain dump + idea library
ideas/[id]/ brain graph, live thinking view, results
settings/ AI provider + profile
api/ideas/... create + reflect, run (streams progress), choose
components/
dashboard/ Overview, Consequences, Validate, Research, charts, graphs
Peep.tsx animated avatar
BrainDots.tsx animated background
lib/
ai/agent.ts the 5-stage pipeline
ai/providers.ts provider, model and key handling
scoring.ts guardrails between the AI and the UI
analysis.ts what-if, simulations, coverage (runs in the browser)
db/schema.ts database tables
Useful commands
npm run dev # start the app locally
npm run typecheck # TypeScript checks
npm run lint # ESLint
npm run test:agent # agent guardrail tests (no API key needed)
npm run build # production build
npm run db:push # create or update database tables
🤝 Contributing
Contributions are very welcome, from typo fixes to new chart types.
- Fork the repo and create a branch.
- Make your change, then run
npm run typecheck && npm run lint && npm run test:agent.
- Open a pull request. CI runs the same checks plus a production build.
Ideas and bug reports are welcome as GitHub issues too.
🗺️ Roadmap
🔒 Privacy
- Your API keys are encrypted at rest and only used on the server to call your chosen provider.
- Your ideas and profile are stored in your own database (the one you deploy), and never shared with anyone else.
- The AI provider you choose receives your idea and profile details to do the analysis. Check their data policy.
- Important life, health, legal or financial decisions deserve a professional too. Connect Brain Dots helps you think; it doesn't replace expert advice.
📄 License
Apache License 2.0. Avatars: Open Peeps by Pablo Stanley (CC0).
Made with 💜 for everyone who has ever overthought lunch.
