docs

Qmmit Documentation

Automatic AI prompt tracking via git hooks. Zero workflow changes.

⚡ 30-Second Quick Start
1Install
curl -fsSL https://aicommit.space/install.sh | bash
2Enable
cd your-project && qmmit init
3Done
git commit & git push work automatically

That's it. Use git commit and git push normally. Prompts are captured and synced automatically.

Table of Contents

1. Installation

No Docker, no Node.js required. One command.

# macOS / Linux
curl -fsSL https://aicommit.space/install.sh | bash
 
# Windows
irm https://aicommit.space/install.ps1 | iex
 
# npm
npm install -g @qmmit/cli

2. Enable Tracking

Run once per repo. Detects your AI tools, installs git hooks, imports past sessions.

cd your-project
qmmit init
 
# Output:
# ✔ Found Claude Code, Cursor, Kiro, Copilot
# ✔ Post-commit hook installed
# ✔ Pre-push hook installed
# ✔ 47 prompts imported from past sessions
# Your heatmap already has data!

Then connect to the platform:

qmmit login

3. Automatic Tracking

After qmmit init, tracking is invisible. Use your normal git workflow:

# Use AI tools normally...
# Then commit as usual:
git commit -m "feat: add auth"
# [qmmit] 3 prompts tracked (claude-code, cursor) → a3f9c12
 
# Push as usual:
git push
# [qmmit] 3 prompts synced → aicommit.space/you

The post-commit hook reads session files from your AI tools, filters by repo path and timestamp, and links prompts to the commit. The pre-push hook syncs to the dashboard. Both are non-blocking — if they fail, git still works normally.

4. Manual Tracking (qmmit add)

For web tools (ChatGPT, Gemini web) that don't leave local session files:

qmmit add "designed the DB schema"
qmmit add "generated test cases" --model chatgpt-4o
qmmit add "debugged race condition" --commit f7e8d9c
 
# Buffered until next git commit, then auto-linked

5. Git Hooks

Qmmit installs two git hooks:

post-commitScans AI tool sessions, matches prompts to commit by timestamp window, stores locally
pre-pushSyncs unsynced prompts to the platform API

Both hooks call qmmit _hook internally. They always exit 0 — never block git. If you use husky or lefthook, Qmmit appends to existing hooks.

6. CLI Commands

Setup

qmmit initOne-time setup: detect tools, install hooks, import sessions
qmmit disableRemove hooks (data stays on dashboard)
qmmit loginAuthenticate with platform
qmmit statusShow tracking status, detected tools, stats

Tracking

qmmit add "text"Manually log a prompt (for web tools)
qmmit add --model chatgpt-4oSpecify model
qmmit add --commit SHALink to a specific past commit
qmmit logView prompt history
qmmit show HEADPrompts matched to last commit
qmmit pushForce sync prompts to platform now
qmmit push --dry-runPreview what will sync

Privacy

qmmit privacyInteractive privacy manager
qmmit scanScan for secrets before push
qmmit redact <id>Redact sensitive content
qmmit set-private <id>Make prompt private

7. Prompt-to-Commit Matching

Prompts are linked to commits using a weighted scoring algorithm:

Temporal Proximity35%

Prompts between previous commit and current commit timestamp

File Overlap30%

Files referenced in prompt match files changed in commit

Keyword Match20%

Prompt words appear in commit message or diff

Code Similarity10%

AI response code appears in git diff

Session Continuity5%

Same session as prior matched prompt

8. Privacy & Security

All data stays local until you push. Secret scanning runs before sync.

qmmit scan          # check for API keys, passwords
qmmit redact <id>   # redact sensitive content
qmmit push --review # approve each prompt before sync

9. Supported AI Tools

Claude Code
Auto — reads ~/.claude/projects/ JSONL files
Cursor
Auto — reads state.vscdb SQLite database
Kiro
Auto — reads workspace-sessions/ JSON files
GitHub Copilot
Auto — reads chatSessions/ JSONL in workspaceStorage
Windsurf
Auto — reads ~/.codeium/windsurf/cascade/
OpenCode
Auto — reads opencode.db SQLite database
Gemini CLI
Auto — reads ~/.gemini/tmp/ session files
ChatGPT / Gemini web
Manual — qmmit add "prompt"

10. Dashboard & Profiles

Your profile at aicommit.space/username shows: AI contribution heatmap, model breakdown, prompt timeline, streak counter, and project-level stats.