RAVEN v0.1 — Pre-Release

RAVEN v0.1 — Initial Pre-Release

⚠️ Pre-release notice: v0.x releases are for testing and evaluation purposes only.
They are not stable releases. Features may change, and breaking changes may occur
between versions. Not recommended for production use.

RAVEN (Retrieval Augmented Adaptive Epistemic Navigation) is a privacy-first, fully offline RAG workspace that runs entirely on your local machine.

What's Included

• 4 main Retrieval Pipelines — Embedded, Hierarchical, Agreement-Based, and Vector-Conditioned, each with Local and Global variants
• Multi-Knowledge Architecture — Organize files into isolated databases to prevent data contamination
• Source Reconstruction — Retrieved answers include highlighted source sections for verification
• Intelligent Ingestion — Automatic file sectioning with summary, keyword, condition, and definition extraction
• Sliding Memory — Long conversation support with SQLite-backed memory recall
• Model Management — Download and switch between Gemma 4 E2B-it, Gemma 4 E4B-it, and Qwen2.5 7B Instruct
• Full Parameter Control — Adjust n_ctx, n_gpu_layers, n_threads per model

Installation

Download RAVEN_V0_1_Setup.exe and run it. No additional dependencies required.

System Requirements

• OS: Windows 10+
• GPU: Vulkan-compatible (recommended)
• VRAM: 6 GB minimum
• RAM: 8 GB minimum, 16 GB+ recommended
• Storage: ~5 GB free

Notes

• Agreement retrieval naming conflict resolved.