v0.2.0

What's New

Trace Sources

• Langfuse connector — read traces from Langfuse (cloud or self-hosted)
• Axiom connector — read traces from Axiom datasets
• Now 5 connectors: JSON, LangSmith, OpenTelemetry, Langfuse, Axiom

Evaluation

• Default model changed to claude-sonnet-4-6 (Anthropic) for better evaluation quality
• Severity floors and structured fix instructions for more actionable reports
• Parallel diagnosis and fix generation for faster runs

DX

• Auto-loads .env files — no need to manually export API keys
• Auto-detects API key from any available provider
• Quickstart hint when CLI is run with no arguments
• --dry-run now shows correct model name and cost estimate

Documentation

• Trimmed README to ~180 lines, moved detailed docs to docs/
• Accurate cost table with 4 model options ($0.04–$0.90 per 10 conversations)
• Full command reference with all options in docs/commands.md
• Balanced comparison table acknowledging competitor strengths

Fixes

• Fixed cost estimates (previously understated by 2–3x for some models)
• Fixed dry-run showing wrong model when using --provider openai
• Removed real customer name from source code comment
• Removed AI assistant config files from repo (.cursorrules, AGENTS.md)

Full Changelog: v0.1.0...v0.2.0