multiview

Multi-model AI code review from your terminal. Six specialist reviewers, zero cost.

$ multiview review app.py

Reviewing app.py with 6 models...

  OK  Security         2.1s  (1 findings)
  OK  Architecture     1.8s  (0 findings)
  OK  Performance      3.2s  (2 findings)
  OK  Edge Cases       2.5s  (1 findings)
  OK  Deep Reasoning   4.1s  (1 findings)
  OK  Code Quality     2.9s  (0 findings)

┌───┬──────┬──────────────────┬──────────────────────────────────────┬────────────────┐
│ # │ Sev  │ Location         │ Issue                                │ Flagged by     │
├───┼──────┼──────────────────┼──────────────────────────────────────┼────────────────┤
│ 1 │ CRIT │ app.py:42        │ SQL Injection via f-string           │ Security,      │
│   │      │                  │ User input interpolated directly     │ Deep Reasoning │
│   │      │                  │ into SQL query. Use parameterized    │                │
│   │      │                  │ queries.                             │                │
├───┼──────┼──────────────────┼──────────────────────────────────────┼────────────────┤
│ 2 │ HIGH │ app.py:15-28     │ Unbounded list growth                │ Performance    │
│   │      │                  │ Results appended in loop without     │                │
│   │      │                  │ limit. Use generator or pagination.  │                │
├───┼──────┼──────────────────┼──────────────────────────────────────┼────────────────┤
│ 3 │ MED  │ app.py:67        │ Missing None check on user.email     │ Edge Cases     │
│   │      │                  │ AttributeError when user has no      │                │
│   │      │                  │ email set. Add: if user.email:       │                │
└───┴──────┴──────────────────┴──────────────────────────────────────┴────────────────┘

 Total: 3 issues  (1 CRITICAL, 1 HIGH, 1 MEDIUM)

How it works

multiview sends your code to 6 free-tier AI models in parallel. Each model reviews from a specialist angle:

Reviewer	Model	Focus
Security	Llama 3.3 70B	OWASP Top 10, injection, auth, secrets
Architecture	Gemma 2 9B	Coupling, cohesion, abstraction, naming
Performance	Mistral Small	O(n²), N+1, allocations, caching
Edge Cases	Gemini 2.0 Flash	Null, boundaries, Unicode, off-by-one
Deep Reasoning	DeepSeek R1	Logic bugs, invariant violations, correctness
Code Quality	Qwen 2.5 Coder	DRY, dead code, idioms, readability

Findings are deduplicated across models, ranked by severity, and show which reviewers flagged each issue.

Install

pip install multiview

Or from source:

git clone https://github.com/JustDreameritis/multiview
cd multiview
pip install -e .

Setup

multiview setup

Walks you through getting free API keys from each provider. Direct links to signup pages. No credit card needed.

Providers used:

• Groq - Security & Architecture
• Mistral - Performance
• Google AI Studio - Edge Cases
• OpenRouter - Deep Reasoning
• HuggingFace - Code Quality

Usage

Code Review

# Review a file
multiview review app.py

# Review a directory
multiview review src/

# JSON output for CI/CD
multiview review app.py --json

# Markdown output for PR comments
multiview review app.py --markdown

# Use specific reviewers only
multiview review app.py --model security --model performance

# Check which reviewers are active
multiview status

Multi-Perspective Questions

# Ask a question to 6 specialist perspectives
multiview ask "What causes inflation?"

# JSON output
multiview ask "Should I use microservices?" --json

Each question is analyzed by 6 perspectives in parallel:

• Factual Analyst — verifies claims, cites established knowledge
• Devil's Advocate — challenges assumptions, finds weaknesses
• Practical Advisor — focuses on what you should actually DO
• Technical Expert — provides deep domain knowledge
• Risk Assessor — identifies what could go wrong
• Synthesizer — merges all views into balanced answer

Output shows each perspective, then a CONSENSUS section with:

• What they agree on
• Where they disagree
• Synthesized answer
• Confidence score (HIGH/MEDIUM/LOW)

Output formats

• Terminal (default): Rich table with colors and severity indicators
• JSON (--json): Machine-readable for CI pipelines
• Markdown (--markdown): Paste into PR comments

Why multi-model?

1. Different models catch different bugs. DeepSeek finds logic errors Claude misses. Gemini catches Unicode edge cases Llama doesn't flag.

2. Consensus = confidence. When 3+ models flag the same issue, you know it's real.

3. Specialist prompts matter. A security-focused system prompt makes the model actually look for OWASP issues instead of generic "improvements."

License

MIT