Repository navigation
What’s New in v2?
🔹 Hybrid Engine Architecture
Moved beyond simple API wrappers. The new engine combines Heuristic Static Analysis (offline, instant) with LLM Intelligence (DeepSeek-V3/R1, LLama models..). Checks for basic logical errors before spending tokens, saving both time and cost.
🔹 Local RAG (Retrieval-Augmented Generation)
Introduced a "Context Manager" that allows dragging & dropping codebase files. The AI now understands your specific project context variable names, architectural patterns, and libraries before suggesting fixes.
Tech Stack: SQLite FTS5 + Vector Embeddings (Hybrid Search).
🔹 Offline & Privacy-First Mode
For sensitive code, the Offline Mode runs purely local checks (Linter + Tree-sitter analysis) without sending a single byte to the cloud. Perfect for enterprise environments or air-gapped coding.
🔹 Structured JSON Outputs
No more parsing regex nightmares. The backend now enforces strict JSON schemas for all AI responses, ensuring reliable integration with other tools and CI/CD pipelines.
🔹 "Pruned" & Polished UI
Took the "less is more" approach. Removed legacy clutter, streamlined the workflow into a single-pane glass design, and added real-time connectivity status. It’s not just powerful; it feels great to use.