Astrai Code Review

AI-powered code review with intelligent model routing. Complex logic goes to powerful models. Formatting and style goes to fast, cheap ones. You save 40%+ without sacrificing quality.

What it does

• Smart routing for reviews: Astrai analyzes the diff complexity and routes to the optimal model. A gnarly concurrency bug gets Opus. A missing semicolon gets Haiku. You only pay for the intelligence you need.
• Structured output: Every review returns typed issues with file, line number, severity (critical/warning/info), message, and a concrete suggestion.
• Strictness modes: Standard catches bugs and logic errors. Strict adds style and best-practice checks. Security mode focuses on vulnerabilities, injection, auth, and data exposure.
• BYOK (Bring Your Own Keys): Your provider API keys stay with you. Astrai decides which model to use, then calls the provider using YOUR key. You pay providers directly.
• Cost tracking: Every review response includes the cost and how much you saved vs always using the most expensive model.
• Local-only mode: If you only set ASTRAI_API_KEY without provider keys, Astrai uses its own hosted models. Still routed intelligently, still cheap.

Setup

1. Get a free API key at as-trai.com
2. Set ASTRAI_API_KEY in your environment or skill config
3. Optionally add provider API keys for BYOK routing (e.g. ANTHROPIC_API_KEY, OPENAI_API_KEY)
4. Run /review on any diff or PR

Usage

/review                     Review the current diff (staged changes)
/review --strict            Strict mode: bugs + style + best practices
/review --focus security    Security-focused review (vulns, injection, auth)
/review --file src/auth.py  Review a specific file

Examples

Basic review of staged changes:

/review

Returns issues found in the current diff with severity levels and suggestions.

Strict review for a PR:

/review --strict

Catches not just bugs but also style violations, naming issues, and missed best practices.

Security audit:

/review --focus security

Focuses on SQL injection, XSS, auth bypass, hardcoded secrets, insecure deserialization, and other vulnerability classes.

Environment Variables