Claude Explain Service

This is a FastAPI-based service that provides AI-powered explanations of compiler assembly output for the Compiler Explorer website. The service uses Anthropic's Claude API to analyze source code and its compiled assembly, providing educational explanations of compiler transformations and optimizations.

Overview

The service is designed to run both locally for development and as an AWS Lambda function via Mangum adapter. It provides intelligent analysis of compiler output, helping users understand how their code translates to assembly language.

For detailed design documentation, see claude_explain.md.

Project Structure

See the source code for the current project structure. Key entry points:

• app/main.py - FastAPI application entry point
• test-explain.sh - Integration test script

Setup

Prerequisites

• Python 3.13+
• uv package manager

Environment Configuration

Create a .env file (NOT in git) with your configuration:

ANTHROPIC_API_KEY=<your-key-here>

# Optional S3 caching configuration could go here
# See "S3 Caching Configuration" section below for details

Install Dependencies

Development

Running Locally

# Start development server
uv run fastapi dev

The service will be available at http://localhost:8000

Testing the Service

# Basic test
./test-explain.sh

# Pretty formatted output
./test-explain.sh --pretty

Running Tests

# Run all tests
uv run pytest

# Run specific test
uv run pytest app/explain_test.py::test_process_request_success

Code Quality

# Run pre-commit hooks (ruff linting/formatting, shellcheck)
uv run pre-commit run --all-files

# Manual linting
uv run ruff check
uv run ruff format

Key Features

Key Features

• Smart assembly filtering for large compiler outputs
• AWS CloudWatch metrics integration when deployed
• Local development with .env file configuration
• S3-based response caching to reduce API costs and improve performance
• Configurable cache with bypassCache option for fresh responses

API Usage

Endpoint

POST / (root path)

Field Naming Convention

The API uses camelCase for all field names to maintain consistency with JavaScript/JSON conventions, even though the internal Python code follows snake_case conventions.

Request Format

{
  "language": "c++",
  "compiler": "g112",
  "code": "int square(int x) { return x * x; }",
  "compilationOptions": ["-O2"],
  "instructionSet": "amd64",
  "asm": [
    {
      "text": "square(int):",
      "source": null,
      "labels": []
    },
    {
      "text": "        push    rbp",
      "source": {
        "line": 1,
        "column": 21
      },
      "labels": []
    }
  ],
  "bypassCache": false  // Optional: set to true to skip cache reads
}

Response Format

{
  "explanation": "The compiler generates efficient assembly...",
  "status": "success",
  "model": "claude-3-5-haiku-20241022",
  "usage": {
    "inputTokens": 123,
    "outputTokens": 456,
    "totalTokens": 579
  },
  "cost": {
    "inputCost": 0.000123,
    "outputCost": 0.000456,
    "totalCost": 0.000579
  },
  "cached": false
}

Configuration

See app/explain.py for current limits and model configuration. The service includes configurable limits for input size, assembly processing, and response length.

Deployment

The service is designed for AWS Lambda deployment with API Gateway. See the Terraform configuration in the repository for infrastructure setup. The version of the service that runs in production is controlled by the terraform in our infra repository.

S3 Caching Configuration

The service supports S3-based response caching to reduce API costs and improve performance. Configure caching through environment variables:

• CACHE_ENABLED: Set to true to enable caching (default: true)
• CACHE_S3_BUCKET: The S3 bucket name for storing cached responses
• CACHE_S3_PREFIX: The key prefix for cache objects (default: explain-cache/)
• CACHE_TTL: HTTP Cache-Control max-age in human-readable format (default: 2d)
  • Examples: 2d (2 days), 48h (48 hours), 30m (30 minutes), 172800s (seconds)
  • Note: This sets the HTTP Cache-Control header for client-side caching only
  • S3 object expiration is handled separately via bucket lifecycle rules

Required S3 Permissions

The Lambda function's IAM role must have the following S3 permissions:

• s3:GetObject - to retrieve cached responses
• s3:PutObject - to store new responses

Objects are stored with an HTTP Cache-Control header for client-side caching. S3 object lifecycle and expiration are managed independently of this TTL.

Cache Behavior

• Cache keys are generated based on the full request content including source code, assembly, and all options
• Requests with bypassCache: true will skip cache lookup but still store the fresh response
• Cache misses or errors gracefully fall back to generating fresh responses