Skip to content

TRACE

TRACE is actionable preventative causal intelligence for chronic disease.

Code Video Cerebral Valley

Core trick

Feeds symptom text through Sonnet parsing, then Opus with extended thinking traces causal chains across body systems using Exa and Firecrawl search.

Tags

health

Description

TRACE is actionable preventative causal intelligence for chronic disease. Instead of telling patients what they already know — that they're in pain — TRACE traces why, connecting symptoms across body systems to their biological root causes. I built this because I have sickle cell disease. A sinoplasty fixed my chronic back pain. No doctor connected sinus inflammation to sickling pain in my spine. TRACE is the tool I needed — one that reasons across systems the way the disease actually moves. Three Claude agents power a real-time pipeline. Sonnet 4.5 parses natural language symptoms. Opus 4.6 with extended thinking traces causal chains across pathophysiology, autonomously pulling evidence through Exa and FireCrawl MCP tools. A second Sonnet 4.5 agent generates urgency-rated, actionable recommendations.
zetsuchan/Trace★ 40 forksTypeScriptMITCreated Feb 14, 2026 · last push Apr 1, 2026

README excerpt

# T R A C E

**Sickle Cell Causal Intelligence** — Trace symptoms to root causes across body systems.

## Why TRACE Exists

I have sickle cell disease. Every time I feel a crisis coming, I'm mentally tracing connections — was it the cold weather? The dehydration? The sinus congestion from last week? No tool does this for me. TRACE does.

TRACE uses multi-agent AI reasoning to trace invisible connections across your body systems, from the symptom you feel to the biological root cause, through every mechanism in between.

## How It Works

Three specialized Claude agents work in sequence:

```
User describes symptoms
  → Symptom Analyzer (Claude Sonnet 4.5)
      Parses input into structured symptoms with body systems and temporal markers
  → Causal Chain Builder (Claude Opus 4.6 + Extended Thinking)
      Traces causal connections through SCD pathophysiology
      Autonomously searches Exa for medical research
      Scrapes articles via FireCrawl when relevant
  → Recommendation Agent (Claude Sonnet 4.5)
      Generates actionable suggestions with urgency ratings
  → Results save to Obsidian vault + PostgreSQL database
```

## Three Visualization Modes

Same trace data, three ways to understand it:

- **Interactive** — Card-based investigation where you click through branching causal paths
- **Timeline** — React Flow node graph showing temporal connections with animated particles
- **Linear** — Step-by-step chain reveal with peeling layer animations

## Key Features

- **Multi-

Team