Independent AI product studio and open research lab.
We build production systems, release open-source tools, and study how agent workflows actually behave.
Kizz is deliberately still taking shape. We are exploring product engineering, agent capabilities, developer infrastructure, and applied AI research instead of forcing everything into one category too early.
The common thread is simple: build real systems, measure what happens, publish what can be shared, and let evidence—including negative and inconclusive results—change the next version.
| Project | Direction | What it is |
|---|---|---|
| Agentic Evidence Lab | Open research | A public method and evidence toolkit for testing versioned skills, prompts, models, tools, councils, and complete agent workflows. |
| Council | Open-source agent capability | Evidence-bound adaptive engineering deliberation for Codex agents, and the first capability developed through Agentic Evidence Lab. |
| Givon AI | Production AI product | An AI content workspace for images, video, creators, agents, and reusable production workflows. |
| Monium CLI | Developer tooling | An unofficial offline-first CLI for Yandex Monium Logs and Traces workflows. |
| Vault MCP Bridge | Agent infrastructure | A self-hosted, read-only Obsidian vault bridge for ChatGPT over MCP. |
| Baron | Product exploration | A peer-to-peer learning platform where people exchange skills using time instead of money. |
We want to move further into open, reproducible research around agent systems:
- which models, tools, skills, prompts, and workflows work for which tasks;
- when multi-agent systems help, when they only add cost, and why;
- how agent capabilities can be tested, compared, and recursively improved;
- how research can produce useful open-source capabilities rather than only papers or benchmark numbers.
There is no final roadmap yet. Kizz is a place for building several credible directions, publishing the strongest ones, and retiring or reshaping the rest.
Meld is paused while its underlying idea—an AI-native relationship with personal knowledge—is being reconsidered. It may return as something broader than the original product.
Evidence > Hype Working systems > Slides
Open when possible Claims stop where proof stops
Est. 2026 · Bootstrapped · Building in public