Conversations don't follow straight lines. You start planning your TikTok schedule, pivot to a song stuck in your head, remember you wanted concert tickets, check your budget, then circle back to content strategy. That's not a flaw — it's how non-linear thinking works.
But it's invisible. You finish a session and wonder: how many topics did I actually touch? Where did I drift? Did I ever come back?
ADHD Drift Visualizer makes the invisible visible. Feed it a chat session and it maps every topic pivot, scores how far you drifted, and shows you the shape of your conversation.
Feed it a conversation. It scores every message for drift from your starting topic and builds a relational graph of how your thinking moved.
graph LR
A["<b>TikTok strategy</b><br>posting schedule"] -->|pivot| B["<b>Song stuck in head</b><br>from a reel"]
A -->|stay| A2["<b>TikTok hashtags</b><br>trending sounds"]
B -->|pivot| C["<b>Concert tickets</b><br>are they sold out?"]
C -->|pivot| D["<b>Budget</b><br>can I afford it?"]
D -->|return| E["<b>TikTok monetization</b><br>creator fund"]
E -->|pivot| F["<b>Instagram Reels</b><br>cross-posting"]
F -->|return| G["<b>TikTok analytics</b><br>back on track"]
style A fill:#39FF14,stroke:#248A3D,color:#000
style A2 fill:#39FF14,stroke:#248A3D,color:#000
style B fill:#FF9F0A,stroke:#C77800,color:#000
style C fill:#BF5AF2,stroke:#8944AB,color:#fff
style D fill:#FFD60A,stroke:#B8A000,color:#000
style E fill:#00E5FF,stroke:#0071A4,color:#000
style F fill:#FF9F0A,stroke:#C77800,color:#000
style G fill:#39FF14,stroke:#248A3D,color:#000
Four signals measure different aspects of each message, then combine into one drift score:
graph TD
MSG["New Message"] --> S1["Semantic Distance<br><i>Global: vs first message</i><br>weight: 0.35"]
MSG --> S2["Topic Divergence<br><i>Local: vs previous message</i><br>weight: 0.35"]
MSG --> S3["Specificity Shift<br><i>Vague or concrete?</i><br>weight: 0.15"]
MSG --> S4["Velocity<br><i>How fast drift accelerates</i><br>weight: 0.15"]
S1 --> SCORE["Composite Score<br>0.0 — 1.0"]
S2 --> SCORE
S3 --> SCORE
S4 --> SCORE
SCORE --> CLASS["Classification"]
SCORE --> GRAPH["3D Topic Graph"]
SCORE --> CURVE["Drift Curve"]
style MSG fill:#58a6b8,stroke:#3d7a8a,color:#000
style SCORE fill:#00E5FF,stroke:#0071A4,color:#000
style CLASS fill:#FF9F0A,stroke:#C77800,color:#000
style GRAPH fill:#BF5AF2,stroke:#8944AB,color:#fff
style CURVE fill:#39FF14,stroke:#248A3D,color:#000
Each message gets classified — no judgment, just a description of distance from the starting topic:
graph LR
subgraph " "
direction LR
T1["ANCHORED<br>0.0 – 0.1"]
T2["BRANCHING<br>0.1 – 0.3"]
T3["TANGENT<br>0.3 – 0.6"]
T4["WANDERING<br>0.6 – 0.8"]
T5["DEEP SHIFT<br>0.8 – 1.0"]
end
T1 --> T2 --> T3 --> T4 --> T5
style T1 fill:#39FF14,stroke:#248A3D,color:#000
style T2 fill:#00E5FF,stroke:#0071A4,color:#000
style T3 fill:#FF9F0A,stroke:#C77800,color:#000
style T4 fill:#FFD60A,stroke:#B8A000,color:#000
style T5 fill:#BF5AF2,stroke:#8944AB,color:#fff
The dashboard shows your initial goal at the top, tracks how far each message drifts from it, and tells you whether you made it back. The topic flow chips, drift curve, and 3D graph all update in real time.
The visualizer doesn't judge. A session with 8 pivots isn't "worse" than one with 0. It's different. The goal is awareness — see your patterns, understand them, and use that knowledge however you want.
Important
This tool is not a diagnostic instrument. It does not diagnose, treat, or assess any medical or psychological condition. It visualizes context-switching patterns in conversations — nothing more.
git clone https://github.com/qinnovates/ADHD-drift-visualizer.git
cd ADHD-drift-visualizer
pip install -e ".[ui,voyage]" # or [ui,openai] or [ui,local]
# Run the demo
adhd-drift demo
# Analyze a chat session
adhd-drift analyze session.md --human
# Launch the dashboard
adhd-drift serve
# Open http://127.0.0.1:3457| Feature | Description |
|---|---|
| 4-Signal Scoring | Multi-signal drift scoring optimized for conversation topic tracking |
| Topic Labeling | Auto-extracts topic labels for each message |
| Pivot Detection | Identifies exact messages where topic shifted |
| Focus Streak | Longest run of anchored/branching messages |
| Topic Flow | Visual timeline of topic changes with color-coded drift |
| Refocus Suggestions | ADHD-aware, non-judgmental suggestions to get back on track |
| Chat Parsers | Import from Claude Code, ChatGPT, or markdown conversation logs |
| Historical Mode | Analyze MEMORY.md files for cross-session patterns |
| 3D Topic Graph | Interactive WebGL graph showing how topics connect and where your thinking steers |
| Dashboard | Real-time visualization with drift curve, 3D graph, topic flow |
| Pluggable Embeddings | Voyage AI, OpenAI, or local — use what you already have |
| CLI | adhd-drift analyze, serve, demo, history — fits into your workflow |
Upload or point at any of these:
| Format | File Extension | Source |
|---|---|---|
| Claude Code JSONL | .jsonl |
~/.claude/history.jsonl |
| ChatGPT Export | .json |
ChatGPT Settings > Data Controls > Export |
| Markdown | .md, .txt |
Any User: / Assistant: formatted conversation |
If you use Claude Code, install the skill for in-session analysis:
# Install the library
pip install -e /path/to/ADHD-drift-visualizer
# Copy the skill
cp -r skills/adhd-drift ~/.claude/skills/adhd-driftThen use it in any session:
/adhd-drift # Analyze the current session
"analyze my drift" # Same thing, natural language
"what do I keep coming back to" # Cross-session memory analysis
"show my topic graph" # Launch 3D interactive graph
Note
The skill asks for consent before reading memory files or conversation history. Your data stays local — nothing is transmitted or persisted beyond the session.
The dashboard includes an interactive WebGL topic graph showing how your thinking steers between subjects:
- Nodes = topics, sized by how often you visit them
- Edges = transitions between topics, with directional particles showing flow
- Colors = drift classification (green=anchored, cyan=branching, orange=tangent, yellow=wandering, purple=deep shift)
- Orbit, zoom, hover for details
Launch it: adhd-drift serve then open http://127.0.0.1:3457 and load a session.
- Self-awareness: See how your conversations actually flow vs how you think they flow
- Session retrospectives: "I touched 7 topics but only completed 2 — what do I prioritize tomorrow?"
- Pair programming: Visualize drift in real-time during a coding session
- Meeting analysis: Score whether a standup stayed focused or went on tangents
Architecture
src/adhd_drift/
├── engine.py # DriftEngine — orchestrates all components
├── scorer.py # 4-signal composite scoring
├── classifier.py # 5-tier classification
├── tracker.py # Cumulative drift + pivot detection
├── recovery.py # Recovery score + refocus suggestions
├── embeddings.py # Pluggable embedding providers
├── signals.py # Specificity shift signal
├── topics.py # Topic extraction + labeling
├── types.py # Shared dataclasses and enums
├── demo.py # Sample session
├── history.py # Cross-session memory analysis
└── parsers/
├── claude.py # Claude Code JSONL
├── chatgpt.py # ChatGPT export JSON
└── markdown.py # Generic markdown
ui/
├── dashboard.html # Single-file dashboard
└── server.py # Flask dev server
Configuration
| Signal | Default Weight | Tune Up For |
|---|---|---|
semantic_distance |
0.35 | Tasks where staying on-topic matters most |
topic_divergence |
0.35 | Detecting subject changes |
specificity_shift |
0.15 | Catching when messages get vague |
velocity |
0.15 | Real-time monitoring, alerting |
| Profile | Max Drift | Alert At | Best For |
|---|---|---|---|
| Strict | 0.15 | 0.10 | Task-focused deep work |
| Standard | 0.30 | 0.25 | Normal conversation |
| Creative | 0.60 | 0.50 | Brainstorming |
| Freeform | 0.80 | 0.70 | Stream of consciousness |
Built by qinnovates