Privacy & Security

Codemetry is designed with privacy as a core principle. Your code stays on your machine.

Core Privacy Principles

1. Local-First Analysis

All analysis runs locally on your machine:

• Git commands execute in your repository
• Signal extraction happens locally
• Scoring and normalization are computed locally
• Results stay on your system

No code, diffs, or file contents are ever sent anywhere.

2. Opt-In AI Only

AI features are disabled by default. When you enable AI:

• Only aggregated metrics are sent (numbers, not code)
• You choose the AI provider
• You control which analyses use AI

3. No Telemetry

Codemetry does not:

• Phone home
• Collect usage statistics
• Track analysis runs
• Send anonymous data

What happens on your machine stays on your machine.

What Data Goes Where

Local Only (Always)

Data	Storage
Source code	Never leaves your repo
Git history	Never leaves your repo
File contents	Never read by Codemetry
Diffs	Never read by Codemetry
Baseline cache	.git/codemetry/ or temp dir
Analysis results	Your terminal/files

Sent to AI Provider (Opt-In Only)

When AI is enabled, these metrics are sent:

Data	Example	Purpose
Aggregated metrics	churn: 450, scatter: 8	Context for explanation
Normalized values	churn_percentile: 95	Show relative significance
Score and label	score: 35, label: bad	What to explain
Confounders	large_refactor_suspected	Additional context
Reasons	High churn at p95	What contributed

Never sent:

• Actual code
• Commit messages (full text)
• File paths (beyond counts)
• Author names/emails
• Repository name/URL

Note

AI receives only what’s needed to explain the numbers. It cannot see your code, cannot identify your project, and cannot identify individuals.

AI Provider Considerations

When using AI, data is sent to your chosen provider:

Provider	Data Handling
OpenAI	OpenAI Privacy Policy
Anthropic	Anthropic Privacy Policy
DeepSeek	DeepSeek Terms
Google	Google AI Terms

Key Points

• API requests may be logged by providers
• Some providers may use data for model training (check their policies)
• Enterprise tiers often offer data retention controls
• Consider using providers with strong data protection commitments

Mitigation

Since Codemetry only sends metrics (not code), even if logged:

• Metrics alone don’t reveal proprietary information
• Numbers can’t be reverse-engineered to code
• No PII is included in the data

Security Considerations

API Key Storage

Store API keys securely:

# Good: Environment variables
export OPENAI_API_KEY=sk-...

# Good: Laravel .env (not committed)
OPENAI_API_KEY=sk-...

# Bad: Hardcoded in config files
'api_key' => 'sk-...',  # Don't do this

Repository Access

Codemetry needs:

• Read access to .git/ directory
• Write access to .git/codemetry/ for caching
• Execute permission for git commands

It does not need:

• Network access (except for AI, which is opt-in)
• Write access to your source files
• Access to other repositories

CI/CD Security

When running in CI:

• Use secret management for API keys
• Consider whether AI summaries are needed in CI
• Cache files don’t contain sensitive data

Compliance Notes

GDPR

• No personal data collection by Codemetry itself
• Git metadata (author names/emails) stays local
• AI providers may have GDPR obligations if you enable AI

SOC 2 / HIPAA

• Local-only analysis supports compliance requirements
• AI integration requires evaluating provider compliance
• Consider disabling AI for sensitive repositories

Internal Policies

If your organization has code security policies:

• Codemetry’s local-only design typically satisfies “code doesn’t leave the network”
• AI opt-in allows per-project decisions
• Audit the metrics sent to AI if needed

Tip

For maximum privacy, simply don’t enable AI. You get full analysis functionality with zero external data transmission.

Verification

Audit the Code

Codemetry is open source. You can verify:

• What data is collected: Check signal providers
• What’s sent to AI: Check AI engine implementations
• Network calls: Search for HTTP client usage

Monitor Network

Run analysis while monitoring network:

# Without AI: No network calls
php artisan codemetry:analyze --days=7

# With AI: Only calls to your chosen provider
php artisan codemetry:analyze --days=7 --ai=1

Review AI Payloads

Enable logging to see exactly what’s sent:

// In a custom AI engine wrapper
Log::debug('AI payload', $payload);

Questions?

If you have privacy concerns not addressed here:

• Review the source code
• Open an issue on GitHub
• Contact the maintainers

We take privacy seriously and welcome scrutiny of our approach.