AI Providers

The tool supports multiple AI providers behind a unified interface. You can switch providers via configuration without changing your workflow.

Supported providers

• openai – OpenAI Chat Completions-compatible models
• gemini – Google Gemini
• anthropic – Anthropic Claude
• ollama – Local models via Ollama
• mock – Deterministic, no-network mock used by default

See src/Providers/* for exact options exposed by each provider implementation.

Selecting a provider

providers:
  default: openai   # or gemini | anthropic | ollama | mock
  # Additional provider-specific keys may be needed (see source code)

Token budgeting

• Approximate token accounting: characters/4.
• Global cap via context.diff_token_limit and per-file cap via context.per_file_token_cap.
• Overflow strategy defaults to trim; other strategies may be added over time.

Advanced Token Optimization

The system includes sophisticated token cost optimization capabilities that can reduce token usage by 30-50% for input and 40-60% for output:

• Semantic Chunking – Groups related code changes by context (classes, methods, etc.) via enable_semantic_chunking
• Diff Compression – Intelligently compresses diffs while maintaining semantic meaning via enable_diff_compression
• Similar Finding Consolidation – Aggregates similar issues across files via consolidate_similar_findings
• Per-file Limits – Controls review scope with max_findings_per_file to prevent overwhelming output
• Severity Limits – Fine-tunes output with severity_limits to cap findings by severity level

Coding guidelines embedding

If guidelines_file is set in config, its contents are base64-embedded into the prompts for all providers. The system instructions direct the model to decode and apply them strictly during review.