kapso

Kapso by Leeroo: an engine that runs experiments toward an objective, deploys the best solution on your stack and learns from every run into a bank of lessons w

What is kapso?

Kapso by Leeroo: an engine that runs experiments toward an objective, deploys the best solution on your stack and learns from every run into a bank of lessons with a credence score and a boundary. First on RelBench and ALE-Bench, top open-source on MLE-Bench, above every human contestant at IOAI 2026.

Overview of kapso

What it is

Kapso is an open-source engine that runs experiments, deploys the best solution, and learns from every run. It starts from a domain harness and an empty bank of lessons. Given an objective and success criteria, it runs a tree of experiments, deploys the best solution on the target stack, and learns from the run by updating the bank of lessons with a credence score and a boundary on where each lesson applies. Kapso is model-agnostic and coding-agent-agnostic, developed by Leeroo, and published on PyPI as leeroo-kapso.

Key features

  • Smart experimentation: each experiment is a hypothesis implemented on its own branch and scored, in a tree. Ideation draws on earlier experiments and the bank of lessons, re-ranking ideas by expected return against budget.
  • Learning: every lesson carries a credence score, a boundary, and evidence. learn() updates lessons from each run's outcome and A/B tests every change by rebuilding with and without it.
  • Deployment: adapts the best solution to the target stack, deploys it with health checks and a working call. Supports Docker, Modal, BentoCloud, and LangGraph Platform.
  • Domain adaptation: harnesses define the knowledge and workflows of an area. Predictive modelling and competitive programming ship; users can write custom harnesses.
  • Model-agnostic and coding-agent-agnostic: works with any model and coding agent CLI.
  • Achievements: first on all three RelBench leaderboards (classification, regression, recommendations), top open-source on MLE-Bench and ALE-Bench, and above every human contestant at IOAI 2026.

Use cases

Kapso is suited for predictive modelling over relational data, competitive programming, algorithmic discovery, Kaggle ML competitions, and AI Olympiad tasks. It can be used for any domain where running experiments toward an objective, deploying the best solution, and accumulating a bank of lessons is valuable.

Getting started

Prerequisites: Node.js 18+, and the coding agent CLIs (@openai/codex and @anthropic-ai/claude-code) must be installed and logged in. Install the Kapso plugin for coding agents: bash claude plugin marketplace add Leeroo-AI/kapso && claude plugin install kapso@kapso codex plugin marketplace add Leeroo-AI/kapso && codex plugin add kapso@kapso Install the Python package: bash pip install leeroo-kapso Verify the setup: bash kapso doctor

License and pricing

MIT license, open source.

Based on the project's public README and repository metadata.

Add this badge to your README

Featured on YunoDB Badge styles & help

Maintain this project? Show that it is featured on YunoDB.

Markdown[![Featured on YunoDB](https://yunodb.com/badge/leeroo-aikapso.svg)](https://yunodb.com/tools/leeroo-aikapso?utm_source=github&utm_medium=badge)
HTML<a href="https://yunodb.com/tools/leeroo-aikapso?utm_source=github&utm_medium=badge"><img src="https://yunodb.com/badge/leeroo-aikapso.svg" alt="Featured on YunoDB" height="20"></a>