Why ClearPR?
The Problem Nobody Talks About
Every team has this workflow:
- Developer finishes a feature
- Runs
npx prettier --write .before committing - Opens a PR
- PR shows 20,847 lines changed
- Reviewer panics, scrolls for 10 minutes, gives up
- Approves with "LGTM" and hopes for the best
The real change? 47 lines. But they're buried in 20,000 lines of Prettier formatting.
GitHub's "Hide whitespace" toggle doesn't help because Prettier changes more than whitespace:
| What Prettier Does | "Hide whitespace" catches it? |
|---|---|
| Rewraps lines at 80 chars | No |
Changes ' to " | No |
| Adds trailing commas | No |
| Reorders imports | No |
| Adds/removes semicolons | No |
| Changes indentation | Yes (only this one) |
5 out of 6 common Prettier changes slip through GitHub's filter.
The Real Diff
Here's what ClearPR shows instead of 20K lines of noise:
What Existing Tools Miss
GitHub Copilot PR Review
- Reviews the raw diff — all 20K lines
- Burns tokens on formatting noise
- Generic advice not tied to your team's rules
- No memory of past reviews
CodeRabbit
- SaaS only — your code leaves your network
- No semantic diff filtering
- No team-specific guideline loading
- No learning from past PR feedback
Graphite / Reviewbot
- Focused on workflow, not review quality
- No AST-based diff filtering
- No formatting noise removal
Manual AI Review (paste diff into ChatGPT)
- Manual process, not automated
- No integration with GitHub
- No project context or guidelines
- Entire raw diff = wasted tokens and poor focus
What ClearPR Does Differently
| Feature | Copilot | CodeRabbit | ClearPR |
|---|---|---|---|
| Strips formatting noise (AST) | - | - | Yes |
| Self-hosted / private | - | - | Yes |
| Your project's guidelines | - | - | Yes |
| Learns from past reviews | - | - | Yes |
| Multi-LLM provider | - | - | Yes |
| Token-efficient | - | - | Yes |
| Advisory only (never blocks) | Yes | Yes | Yes |
Sample Review Output
When ClearPR reviews a PR, it posts:
Inline Comments
Summary Comment
The summary shows:
- Diff stats — how many lines were noise vs real changes
- Findings grouped by severity (critical, warning, info)
- Details table with file, line, and issue description
- Whether guidelines and past feedback were used
The Math: Token Savings
By filtering formatting noise before sending to the LLM:
| PR Size | Raw Tokens | ClearPR Tokens | Monthly Cost (team of 10) |
|---|---|---|---|
| Small (50 lines) | ~2,000 | ~1,500 | - |
| Medium + formatter | ~15,000 | ~4,000 | - |
| Large + prettier | ~80,000 | ~8,000 | - |
| Total (200 PRs/mo) | ~$80-220 | ~$25-60 | 60-73% savings |
Who Should Use ClearPR
- Teams running Prettier/ESLint — the #1 source of noisy diffs
- Teams with documented coding standards — ClearPR enforces them
- Security-conscious orgs — self-hosted, no code leaves your network
- Cost-conscious teams — dramatically reduces LLM token usage
- Anyone tired of scrolling through 20K-line PRs to find 47 real changes