Draw your thought process as a directed graph. ThoughtFlow researches every node with real-time web intelligence and propagates compounding context through your connections — just like a neural network.
Your thought structure becomes a computational graph. AI researches each node and compounds the knowledge downstream.
Drag 9 node types onto the infinite canvas. Connect them with typed edges — dependencies, contrasts, causality. Structure your thinking visually.
Your graph transforms into an immersive neural network. Glowing spheres pulse as each node is researched. Context flows through edges as particle streams.
Downstream nodes receive the full accumulated context of every ancestor — getting progressively richer and more nuanced as research flows through your graph.
A structured research report with per-node findings, alternate simulation paths, and a final synthesized recommendation. Cite sources, export, and share.
Everything you need to externalize, research, and synthesize complex thinking.
When you hit Run, your 2D graph morphs into a Three.js neural network. Nodes become glowing spheres, edges become energy beams with flowing particle systems. The hero moment.
Three.js + React Three FiberEach node's research output is distilled and pushed downstream — exactly like activations through a neural network. Deep graphs produce exponentially richer outputs.
Topological executionEvery node generates optimized search queries and retrieves current web results via Tavily. No stale training data — your research is always fresh.
Tavily Search APIAfter your primary simulation completes, ThoughtFlow generates 2–3 alternate paths — identifying assumptions you missed, perspectives you didn't consider, and risks you didn't model. Your thinking, stress-tested by AI.
Claude SonnetResults stream in word-by-word via SSE as AI researches each node. Watch your report write itself in real time — no waiting for a bulk dump.
Server-Sent EventsDrag PDFs, Word docs, and spreadsheets onto datapoint nodes. ThoughtFlow extracts and incorporates your own documents into the research context.
PDF · DOCX · XLSXPlug in your own API keys for Anthropic, OpenAI, Groq, or Gemini. Choose the model that fits your budget and precision needs per execution.
4 providers supportedExport your full research report as PDF or Markdown. Generate a unique share link for each execution run. Send your thinking to anyone.
PDF · MD · Share linkEvery graph is saved. Return, re-run, or fork previous work. Start fast with built-in templates: Decision Tree, Research Hierarchy, Investment Thesis, SWOT.
Auto-save · localStorageEach node type triggers a different research behavior in the AI pipeline.
Web search + synthesis → direct answer with sources
Finds supporting and contradicting evidence, returns a validation score
Structured overview with sub-themes and key insights
Searches for current stats, returns verified figures
Evaluates options against upstream context, returns comparison matrix
Filters downstream research to respect defined limits
Generates expansions, analogies, and alternatives — optimized for divergent output
No independent research — purely merges and synthesizes upstream outputs
Terminal node — generates the final formatted deliverable from everything upstream
When you click Run Simulation, the 2D canvas dissolves. Every thought becomes a glowing sphere. Every connection becomes a pulsing energy beam. The camera orbits your mind as AI researches each layer in real time.
Every node goes through the same rigorous pipeline before its output is passed downstream.
Collect all upstream node outputs. Apply edge weights — high-weight edges contribute more to the merged context. Prepend user-defined global context if set. This becomes the "hidden state" the node receives before any research begins.
Python weighted merge · 2000 token capThe node prompt + assembled context is sent to Claude Haiku. It generates 2–4 diverse, specific search queries designed to maximize coverage and avoid redundancy.
Claude Haiku · query diversity optimizationEach query is executed asynchronously via Tavily Search. Top 3 results per query are collected and deduplicated by URL. Sources are scored for quality and recency.
Tavily API · async batch · deduplicationClaude Sonnet receives the node prompt, upstream context, and all search results. It synthesizes (not repeats), cites every claim, assigns a confidence score, and formats the output according to node type.
Claude Sonnet · structured synthesis · confidence 0–100%The full research output is distilled into a concise context payload — max 500 tokens — containing key findings, confidence level, and open questions. This payload flows to all downstream nodes.
Claude Haiku summarization · 500 token emissionThoughtFlow is domain-agnostic. Wherever complex thinking happens, ThoughtFlow can research it.
Build the logical structure of your investment case as a graph. Each node researches its piece — market dynamics, financials, risks, policy outlook — and passes compounding context to the final Decision node, which synthesizes everything into a clear recommendation.
Try this template →Research what's trending, what competitors are covering, then generate differentiated creative angles that combine both into a unique hook. The Output node synthesizes everything into a complete script outline ready to record.
Try this template →Define your constraints as dedicated nodes. The Decision node automatically receives your team size, budget, and technical research — generating a recommendation that's grounded in your actual context, not generic best practices.
Try this template →Map every dimension of a complex personal decision as nodes. ThoughtFlow researches each one and synthesizes a nuanced recommendation that weighs all your constraints — not just the easy-to-Google parts.
Try this template →Survey the existing literature, validate your hypothesis against it, gather the key data points, and synthesize them into a polished research abstract — all from a single graph execution with real cited sources.
Try this template →ThoughtFlow treats your thought graph as a computational graph — the same mathematical structure that powers modern AI.
No credit card required. Free tier gives you real executions — not just a demo.
Build your first thought graph in minutes. ThoughtFlow researches every node, compounds the context, and synthesizes a report that follows your exact reasoning structure — not a generic AI answer.