v0.2.0 — Initial public release
What's new in v0.2.0
First public release of ContextOps — cache-aware prompt structure optimizer with LLM-as-judge eval.
Highlights
- 🚀 Cache-aware reordering — moves stable sections (system, tools) to the top, variable (query, history) to the bottom. Same total tokens, +3.9% to +13% cache hit rate in our internal bench.
- 📊 LLM-as-judge eval — built-in metrics: faithfulness, relevance, completeness, conciseness.
- 🧪 A/B testing — run two prompts over a golden dataset, get structural + quality deltas.
- 💾 Local SQLite logger — every LLM call logged to
~/.contextops/calls.db. Zero cloud. - 🛠️ Bench harness — 1000+ prompts through Ollama, LM Studio, or OpenRouter.
- 🎯 Acceptance criteria — 30+ formal pass/fail criteria in
docs/ACCEPTANCE.md. - ✅ 39 unit tests — full unit-test coverage, <2s test suite.
Install
pip install contextops
# with LiteLLM auto-callback:
pip install "contextops[integrations]"