Introduction

Welcome to cozymori

cozymori is an organization building open-source AI observability tools. Our mission is to make monitoring, caching, and healing AI functions feel intuitive and — dare we say — cozy.

Ecosystem

VectorWave

A decorator-based Python framework that vectorizes function calls and stores them in Weaviate. VectorWave provides semantic caching, self-healing, distributed tracing, and drift detection for AI/LLM applications.

pip install vectorwave

Key features:

• @vectorize decorator for tracing and caching any function
• Semantic caching — similar inputs return cached results instantly
• Self-healing — AI analyzes errors, generates fixes, opens GitHub PRs
• Distributed tracing with trace_id/span_id hierarchy
• Semantic drift detection with webhook alerts
• RAG-powered code Q&A over your codebase
• Rust-accelerated batch writing via PyO3

VectorSurfer

A full-stack observability dashboard for VectorWave. Built with Next.js 16 + FastAPI, VectorSurfer provides real-time monitoring, trace visualization, AI-powered error diagnosis, and replay testing.

Open VectorSurfer →

Key features:

• Bento grid dashboard with drag-and-drop widgets
• Distributed trace waterfall visualization
• AI Healer — GPT-4 powered error diagnosis (single and batch)
• Replay testing with exact match and semantic comparison
• Hybrid semantic search + RAG Q&A
• Multi-language UI (English, Japanese, Korean)

VectorSurferSTL

A lightweight Streamlit-based alternative to VectorSurfer. Same core monitoring features in a single-process Python app — ideal for quick setups and local development. Open source under the MIT License.

pip install vectorsurferstl
vectorsurferstl

VectorCheck

A CLI regression testing framework for AI/LLM applications. Traditional assert a == b fails for generative AI — VectorCheck uses vector similarity and LLM judge evaluation instead.

pip install vectorcheck

# Run regression tests
vw test --target all --semantic --threshold 0.85

Testing modes:

• Exact match — compare outputs literally
• Vector similarity — compare via embedding cosine similarity
• LLM Judge — use GPT-4 to evaluate semantic equivalence

Architecture

┌──────────────────────────────────────────┐
│              VectorWave                   │
│  (Core SDK — traces, caches, heals)      │
│  Stores execution data in Weaviate       │
└──────────┬───────────┬───────────┬───────┘
           │           │           │
           v           v           v
┌──────────┐ ┌────────────┐ ┌───────────┐
│VectorSurfer│ │VectorSurfer│ │VectorCheck│
│(Next.js +  │ │STL         │ │(CLI Test  │
│ FastAPI)   │ │(Streamlit) │ │ Framework)│
└────────────┘ └────────────┘ └───────────┘

Philosophy

• Developer First — APIs designed by developers, for developers
• Zero Bloat — One decorator does it all, no configuration bloat
• Python Native — Built for the Python AI ecosystem with Rust for performance
• Open Source — MIT licensed, community-driven

Requirements

• Python 3.10 — 3.13
• Docker (for Weaviate vector database)
• OpenAI API key (for self-healing and RAG features)

Next Steps

• Read the Quick Start guide to get up and running
• Explore the VectorWave and VectorSurfer product pages