Files
kata/MODEL_CARD.md

2.0 KiB

language, license, library_name, pipeline_tag, tags, metrics, model-index
language license library_name pipeline_tag tags metrics model-index
en
code
proprietary hanzo reinforcement-learning
autonomous-agent
self-healing-ci
code-generation
agentic-ai
api-only
swe_bench_verified
devbench_pass_rate
repair_iterations
name results
hanzoai/kata
task metrics
type name
software-engineering SWE-bench Verified
type value name
pass@1 0.492 Issue Resolution Rate
task metrics
type name
autonomous-execution DevBench
type value name
pass@1 0.724 Full-Stack Task Pass Rate

Hanzo Kata: Autonomous Agent & Self-Healing Execution System Card

⚠️ Platform Runtime · Closed Source
Hanzo Kata is the autonomous task DAG and self-healing software engineering agent runtime.


1. System Overview

Kata is an autonomous agent framework built on Hanzo's dual decision-generative plane:

  • Decision Engine (Kai): Evaluates preflight safety, tool selection, risk triage, and termination checks at every step ($0.021/M tokens).
  • Reasoning Engine (Zen / Enso): Synthesizes minimal unified diff patches and generates code solutions.
  • DAG Execution Loop: Ingests compiler/test feedback, iterates over repair candidates, and halts on deterministic pass criteria.

2. Verified Autonomous Agent Benchmarks

Evaluated on industry-standard benchmark suites:

Benchmark Task Focus Kata Agent Pass@1 Human Engineer Baseline Standard LLM (Zero-shot)
SWE-bench Verified Real GitHub Bug Fixes 49.2% 62.0% 19.5%
DevBench Multi-file Software Projects 72.4% 81.0% 34.0%
Self-Healing CI Loop Automatic Compiler Bug Fix 84.6% (< 3 turns) — 41.2%