ZivK00 Labs

AI systems · developer tooling · evaluation

I build the infrastructure behind AI agents.

Two open-source tools and one prototype. One decides how work runs and refuses what is unsafe. One decides what the model sees, then measures whether that choice beat a plain keyword search. The third builds an editable first cut from raw footage.

See the work

Systems

Three systems, three proofs.

Open one to see how it works, and where it broke.

01

Routing · guardrails · receipts

AgentLane

Open source · MIT

Decides which tool is allowed to carry out an AI task, refuses the ones that have never actually worked, and keeps a record of what really ran.

Read the source on GitHub

02

Context selection · budgets · benchmarks

ContextForge

Open source · MIT

Picks which files an AI should read when it cannot read everything, then checks whether that choice really beat a plain keyword search.

Read the source on GitHub

03

Video pipeline · local rendering

Rushly

Prototype

Builds a first cut from your clips, photos and a music track — cut, ordered and exportable on your own machine.

Ask for access

Other work

Delivered, and still online.

Client and product work. It proves delivery; it is not what I want to be measured on.

  • The Sites Clients screen. The live link still serves the previous version.
    Local business sites

    Sites Clients

    A clear local offer, mobile-first, deployed and reachable.

  • Mobile game

    Focus Empire

    A focus session visibly builds a city. Playable, not published.

    In constructionAsk for the demo

Contact

A system to make controllable?

Open to engineering roles, missions and collaboration.

vlad@zivk00-labs.com
  • AI systems
  • Developer tooling
  • Evaluation
  • Open source