New: AI Spend Analytics — track token usage, model costs, and tool drift across all your AI toolsLearn more →
Prompt tracking · Token analytics · 7 AI tools

Your prompts
ship code.
Now there's proof.

Qmmit links every AI prompt to the commit it produced. Track token spend, model usage, and tool drift — automatically via git hooks.

Get started — free
Also on·All options →
AI Contribution HeatmapToken tracking · Model spend · Tool visibility
Less
More
terminal
# one-time setup
$ qmmit init
✔ 3 tools · hooks installed
# then just git
$ git commit -m "feat: auth"
[qmmit] 3 prompts → a3f9c12
$ git push
[qmmit] synced → dashboard

Works with the tools you already use.

Claude CodeCursorKiroGitHub CopilotWindsurfOpenCodeGemini CLI

No plugins. No config. Git hooks handle everything.

2.1M+
Prompts Tracked
👤
12,400
Developers
🤖
15+
Models Supported
🔗
890K
Commits Linked
Core Feature

AI Score — Your Verified AI Fluency Metric

A composite 0–100 score that proves how effectively you leverage AI tools in real development workflows. Not self-reported. Not gameable. Verified through prompt-to-commit linkage.

87Expert

What employers see on your profile

A single number that answers: "Can this developer ship with AI?"

6 Weighted Factors

Each factor uses logarithmic curves and anti-gaming measures. The score decays naturally — you have to keep building.

25%
Verification

Prompts that shipped as real commits with HIGH confidence matching

20%
Consistency

Active coding days over 90-day window — sustained practice, not bursts

20%
Model Fluency

Multi-model diversity + intentional model selection per task

15%
Volume

Total verified prompts with logarithmic scaling — prevents gaming

10%
Breadth

Distinct projects with AI-linked commits — cross-domain application

10%
Efficiency

Prompts-per-commit ratio — fewer focused prompts = better engineering

Tier Badges

Elite90–100
Expert75–89
Proficient50–74
Developing25–49

For Developers

Prove your AI fluency with verified data. Stand out in job applications with a score employers trust.

For Employers

Filter candidates by real AI capability. See who ships code with AI vs. who just chats with it.

Anti-Gaming

Logarithmic scaling, efficiency penalties for spam, 90-day recency window. You earn it by building.

Your AI portfolio, auto-generated

Every prompt you write becomes a verified portfolio artifact. Here's what your public profile looks like.

qmmit.dev/alexmorgan
👨‍💻

alexmorgan

✔ CLI VerifiedPro

Full-stack developer · AI-native builder · Open source contributor

1,240 prompts23 projects847 commits12 workflows
91% avg confidence
across all matched prompts
AI Models Used
Claude 3.5 Sonnet48%
GPT-4o31%
GitHub Copilot21%
Prompt Activity
Last 13 weeks
Top Tags
#auth#refactor#api#testing#typescript#security#performance#database
Pinned Projects
saas-starterTypeScript

Full-stack SaaS boilerplate with auth, billing, and teams

312 prompts🔗 89 commits
ClaudeGPT-4o
ai-search-enginePython

Semantic search over docs using PG Vector + OpenAI embeddings

187 prompts🔗 54 commits
GPT-4oCopilot
Recent Prompts
a3f9...claude-3-5-sonnet2h ago
HIGH 91%
Refactor auth middleware to use JWT tokens instead of sessions, keep backward compat
feat: JWT auth
#auth#security
b7c2...gpt-4o5h ago
HIGH 84%
Write comprehensive unit tests for the payment service including edge cases and mocks
test: payment service
#testing
c1d8...github-copilot1d ago
MEDIUM 67%
Add rate limiting middleware to all public API endpoints using Redis
feat: rate limiting
#api#performance

AI Spend Visibility

Know where your tokens go.

Most teams have no idea which AI tool consumes the most tokens, which model drifts in quality, or where their spend actually lives. Qmmit makes it visible.

Total Tokens
2.4M
1.8M in · 620K out
Est. Spend
$12.40
This month
Models Used
6
Across 3 tools
Active Days
18
Last 30 days

Model Usage & Cost

claude-3.5-sonnet
142 prompts1.2M$5.40
cursor/auto
89 prompts680K$3.20
gpt-4o
34 prompts420K$2.80
gemini-pro
12 prompts98K$0.60

AI Tool Breakdown

Cursor
142 prompts1.1M
Kiro
89 prompts820K
GitHub Copilot
34 prompts380K
Manual (ChatGPT)
12 prompts98K
📊

Token-level visibility

See exactly how many input and output tokens each model consumes. Per project, per tool, per day. No more guessing where your AI budget goes.

💰

Estimated spend tracking

Automatic cost estimation based on public model pricing. Know your monthly AI spend without checking 5 different billing dashboards.

🔄

Tool drift detection

Track which AI tools you use over time. Spot when a model switch changes your output quality. See weekly trends across all your projects.

"Most teams have no idea where their AI spend or signal actually lives. Visibility first, optimization second — that's the order most builders skip."

How it works

3 steps. Zero effort.

You code with AI like you always do. Qmmit runs silently in the background and builds your portfolio automatically.

1

Install once

One command in your terminal

$ curl -fsSL qmmit.dev/install.sh | bash

Downloads a small tool called qmmit. Works on Mac, Linux, and Windows. Takes about 30 seconds.

2🔗

Set up your repo

Run once per project

$ cd my-project && qmmit init

Qmmit installs invisible hooks into your git repo. From now on, every commit automatically captures your AI prompts.

3🚀

Just code normally

Nothing changes about how you work

$ git commit && git push

Use Cursor, Copilot, Claude, or any AI tool. When you commit and push, your prompts sync to your profile automatically.

What actually happens?

Here's the full picture, explained simply.

💬
You·Ask your AI tool something
"Add a login form with email and password validation"

You type a prompt in Cursor, Copilot, Claude Code, or any supported tool.

🤖
AI tool·Writes the code and saves a session file
Creates auth.tsx with form validation logic

Every AI tool saves a local log of your conversation. Qmmit reads that log — it never intercepts your network traffic.

📝
You·Commit your changes
git commit -m "feat: add login form"

The moment you commit, a tiny script fires automatically. It reads the AI session log, finds your prompts, and links them to this commit.

🔒
Qmmit·Stores everything locally first
Saved to .qmmit/history.db on your machine

Nothing leaves your computer yet. Your prompts are stored in a local database. You control what gets shared.

📤
You·Push to GitHub
git push

When you push, Qmmit automatically syncs your prompts to your profile. Secrets are auto-redacted before anything is sent.

🌐
Your profile·Updates instantly
qmmit.dev/yourusername

Your heatmap, model breakdown, and prompt timeline update. Anyone can see your verified AI contribution history.

Do I need to change how I code?

No. Use your AI tools exactly as you do today. Qmmit works in the background.

Is my code or prompts sent anywhere without my permission?

Nothing leaves your machine until you run git push. You control what syncs.

What if I use multiple AI tools?

Qmmit supports 7 tools: Cursor, Copilot, Kiro, Claude Code, Windsurf, OpenCode, and Gemini CLI. All captured automatically.

What if I use ChatGPT or Claude.ai (web)?

Those don't save local files, so use: qmmit add "your prompt" to log them manually.

Does it work with private repos?

Yes. Private repos stay private — prompts are hidden from your public profile automatically.

What does "verified" mean?

Every prompt is linked to a real git commit SHA. Anyone can verify it exists on GitHub/GitLab/Bitbucket.

Get started — free

No credit card. No config files. Works in 60 seconds.

GitHub shows what you shipped.
Qmmit shows how you think.

92% of developers use AI tools daily. 41% of code is AI-generated. Job posts asking for AI experience grew 340% in one year. But there's no standard way to prove AI fluency.

🗂️
GitHub / GitLab / Bitbucket
What recruiters see today
Commit history 847 commits this year
Languages TypeScript, Python, Go
Repos 23 public repositories
AI tool usage Unknown — not tracked
Which AI models No data
Prompt quality No data
AI contribution % No data
Verified vs claimed Self-reported only
>_
Qmmit Profile
Verified AI attribution
Commit history847 commits this year
LanguagesTypeScript, Python, Go
Repos23 public repositories
AI tool usage1,240 prompts tracked
Which AI modelsClaude 48%, GPT-4o 31%, Copilot 21%
Prompt qualityTagged, rated, community-forked
AI contribution %Per-project breakdown
Verified vs claimedCryptographically verified

Everything to prove you build with AI.

Capture prompts. Link them to commits. Build a verified portfolio.

⚡ Core Feature

Autocapture from 7 AI tools

Git hooks read session files on every commit. Claude Code, Cursor, Copilot, Kiro, Windsurf, OpenCode, Gemini CLI. You change nothing about how you work.

  • Reads session files from Claude Code, Cursor, Kiro, GitHub Copilot, Windsurf, OpenCode, Gemini CLI
  • Git hooks fire silently — never blocks git operations
  • Matching algorithm links prompts to commits by timing + files
  • All data stays local until git push syncs to dashboard
qmmit log — prompt history
a3f9...claude-3-5-sonnet
Refactor auth middleware to use JWT tokens instead of sessions
feat: JWT authHIGH 91%auth.ts, middleware.ts
b7c2...gpt-4o
Write unit tests for the payment service with edge cases
test: payment serviceHIGH 84%payment.test.ts
c1d8...github-copilot
Add rate limiting to the API endpoints
feat: rate limitingMEDIUM 67%api/routes.ts
Matching algorithm — confidence breakdown
File Overlap35%
auth.ts, middleware.ts matched
Temporal Proximity25%
12 min before commit
Keyword Match20%
"JWT", "auth" in commit diff
Code Similarity15%
Response code found in diff
Session Continuity5%
Same qmmit session
Total confidence91% — HIGH

Every prompt linked to a commit

A 5-signal scoring algorithm matches prompts to commits by file overlap, timing, keywords, code similarity, and session continuity. No manual tagging required.

≥ 80%
HIGHAuto-matched, published immediately
50–79%
MEDIUMYou confirm before publishing
< 50%
REVIEWQueued for manual review
📋

Your prompts are portfolio pieces

Each prompt displays the model, files touched, confidence score, and the commit it produced. Pin your best work. Share individual prompts with a link.

Prompt #a3f9claude-3-5-sonnet · temp 0.7
"Refactor auth middleware to use JWT tokens"
#auth#refactor#security
→ feat: JWT auth (HIGH 91%) · 2 forks · 14 views
🔍

Which AI wrote which code

Per-project model breakdown. Claude 48%, GPT-4o 31%, Copilot 21%. Color-coded attribution bars on every project card.

my-saas-project — AI model breakdown
Claude 3.5 Sonnet48%
GPT-4o31%
GitHub Copilot21%
🔒

Your prompts never leave your machine until you push

Local SQLite database. AES-256 encryption at rest. Auto-redacts API keys, tokens, passwords, and PII before sync. Public, private, or redacted per prompt.

🔐
1. Local only
SQLite on your machine. Zero network until qmmit push.
🔐
2. Encrypted
AES-256 at rest. TLS 1.3 in transit. Keys never leave your device.
🔐
3. Granular control
Public / Private / Redacted per prompt. Auto-scan for API keys, PII, passwords.
🔗

GitHub / GitLab / Bitbucket

OAuth import from all three platforms. Retroactive AI analysis on existing repos. Auto-detected from .git/config.

🐙
GitHub
OAuth App
Webhooks
Actions CI
Enterprise
🦊
GitLab
OAuth 2.0
Webhooks
CI/CD
CE/EE
🪣
Bitbucket
OAuth 2.0
Webhooks
Pipelines
Server
🛠️

AI Workflow Builder

Document your AI-assisted development process as a reusable, shareable workflow. Link each step to the prompts and commits that produced it. Fork and remix workflows from other developers.

  • Visual step-by-step editor — drag, drop, reorder
  • Link steps to specific prompts and commits
  • Share publicly or keep private
  • Fork and remix from the community
  • Export as markdown or embed in README
Try Workflow Builder →
Example: "Build a REST API with AI" workflow
1
Design the schema
"Design a PostgreSQL schema for a SaaS app with users, teams, and billing"
claude
2
Generate the models
"Generate TypeScript Prisma models from this schema"
gpt-4ofeat: db schema
3
Write the routes
"Create Express routes for CRUD operations on the User model"
copilotfeat: user routes
4
Add auth middleware
"Add JWT auth middleware that validates tokens and attaches user to req"
claudefeat: auth middleware

Full CLI reference

Every git command works. Plus AI tracking on top.

Setup
qmmit init
Detect tools, install hooks, import sessions
qmmit login
Connect to qmmit.dev
qmmit disable
Remove hooks (data stays)
qmmit status
Tracking status + stats
Manual Tracking
qmmit add "Fix auth bug"
Log a prompt (web tools)
qmmit add --model=chatgpt
Specify model
qmmit add --commit SHA
Link to past commit
qmmit log --last 20
View recent prompts
Sync & View
qmmit push
Force sync prompts now
qmmit push --review
Approve each prompt
qmmit push --dry-run
Preview what syncs
qmmit log --model=claude
Filter by model
Inspect
qmmit show HEAD
Prompts for last commit
qmmit show <sha>
Prompts for any commit
qmmit tag <id> arch
Tag a prompt
qmmit note <id> "text"
Add a note
Privacy
qmmit privacy
Interactive privacy manager
qmmit scan
Scan for secrets / API keys
qmmit redact <id>
Redact sensitive content
qmmit set-private <id>
Mark prompt private
Automatic (via git hooks)
git commit -m "feat: x"
Hook auto-captures AI prompts
git push
Hook auto-syncs to dashboard
qmmit _hook post-commit
Internal: scan + ingest
qmmit _hook pre-push
Internal: sync to API
Works with any AI tool — log prompts from all of these
ChatGPT
Web & API
Claude
Web & API
Gemini
Web & API
GitHub Copilot
IDE
Cursor IDE
IDE
Kiro
IDE
Windsurf
IDE
Aider
CLI
Claude Code
CLI
Ollama
Local
LM Studio
Local
Any AI tool
qmmit prompt