Using the ChatPhone a Friend

Phone a Friend (Peer Review)

Get a second AI to review the first AI’s work. Because even AIs need fact-checkers.


⚠️ Note: This feature is currently under active development. The UI is in place but the API route is being refined. Check back soon!


What Is It?

When Claude gives you an answer, you can ask GPT to review it. Or vice versa. A different AI model looks at the first one’s work and calls out:

  • Issues — Things that are wrong
  • Suggestions — Things that could be better
  • Confirmations — Things that are correct

It’s like having two experts on call instead of one.


How to Trigger It

Manual Trigger

Hover over any AI response and click ”🤝 Get Second Opinion”.

Auto-Review (Coming Soon)

Enable in Settings to automatically review important responses.


Review Focus Areas

When triggering, you can pick a focus:

FocusGood For
CodeSyntax, bugs, best practices
SecurityVulnerabilities, auth issues
ArchitectureDesign patterns, scalability
AccuracyFactual correctness
GeneralBroad review, all areas

Reading the Review

Reviews appear below the original response:

┌────────────────────────────────────────┐
│  🤖 Claude's Response                  │
│  [Original answer here]                │
├────────────────────────────────────────┤
│  🔍 Review by GPT-4                    │
│                                        │
│  ⚠️ CRITICAL: SQL injection risk in    │
│     the query builder                  │
│                                        │
│  💡 SUGGESTION: Consider using         │
│     parameterized queries              │
│                                        │
│  ✅ CONFIRMED: Auth flow is correct    │
└────────────────────────────────────────┘

Severity Levels

IconMeaning
🔴 CriticalSerious problem, fix immediately
🟡 WarningPotential issue, worth addressing
🔵 SuggestionImprovement idea, not urgent
ConfirmedReviewer agrees with this point

Cross-Provider Review

The magic: the reviewer is always a different provider.

If Original Is…Reviewer Is…
ClaudeGPT or Gemini
GPTClaude or Gemini
GeminiClaude or GPT

This reduces blind spots. Different models have different training data and biases.


Memory Writeback

When a review catches something useful, Chorum can save it to project memory:

  1. Click “Save Pattern” on a review item
  2. It becomes a learned pattern
  3. Future responses will consider it

Example: If GPT catches that Claude forgot input validation, that pattern gets saved. Next time you ask about forms, the AI remembers.


Cost

Reviews cost money — you’re making an additional API call. The cost appears next to the review:

Review: $0.004

For code review and security checks, it’s usually worth it.


When to Use It

Yes, use it for:

  • Production code
  • Security-sensitive stuff
  • Important decisions
  • Anything you’ll ship

Maybe skip it for:

  • Quick questions
  • Learning/exploration
  • Fun conversations
  • Local models (slower)

Force a Reviewer

By default, Chorum picks the reviewer. But you can force one:

  1. Click “Get Second Opinion”
  2. Hold Shift while clicking
  3. Pick your preferred reviewer

Useful if you specifically want Claude’s take on GPT’s code, or vice versa.


Back to: Chat Overview