# hexo-ai/sia

SIA is a Self Improving AI framework to autonomously improve the performance of any AI system (Model / Agent) on a benchmark task.

Repository: https://github.com/hexo-ai/sia
Canonical: https://ross.abutalabs.com/products/hexo-ai-sia
Homepage: https://hexolabs.com/
Language: Python
License: MIT
License Family: permissive
Last push: 2026-08-26T17:02:30+00:00

## Health v2 (maintenance only)
Score: 59/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 99, release rhythm 35, longevity 11
- inputs: {"age_days": 161, "days_push": 7, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2123, forks 254 (observed 2026-08-28T04:06:16.748461+00:00)

## What it is
SIA is a Python framework implementing a self-improving AI loop in which a Meta-Agent generates a task-specific Target Agent and a Feedback Agent iteratively updates both the agent's harness and model weights to improve benchmark performance. It is the official open-source implementation of the SIA paper (Hebbar et al., 2026) and reports state-of-the-art results on benchmarks like MLE-Bench, LawBench, and CUDA kernel optimization.

## Use cases
- automatically improve an AI agent's performance on a benchmark task
- run a self-improving loop that updates agent harness and weights
- optimize LLM agents for Kaggle-style ML competitions
- improve model performance on scientific tasks like RNA denoising
- generate and refine task-specific agents autonomously
- benchmark self-improving agent systems

## When to choose
- you want an agent system that autonomously improves itself over successive generations
- you need to optimize a model or agent against a measurable benchmark task
- you are researching self-improving or meta-learning agent architectures
- you want to reproduce or build on the SIA paper's results

## When to avoid
- you need a simple, static agent pipeline without iterative self-modification
- you lack the compute budget for repeated training and evaluation loops
- you need a production-ready, battle-tested framework rather than a research system
- your task has no clear automated evaluation signal for the feedback loop

## Facets
- artifact type: framework
- maturity: active
- function: agent-framework, llm-training, machine-learning, benchmarking
- domain: artificial-intelligence, machine-learning, large-language-models
- platform: python
- tags: self-improving-ai, meta-agent, feedback-loop, agent-optimization, llm-agents, autonomous-agents, research, ai-agents, linux, macos, gpu

## Member repositories
- hexo-ai/sia (main) score 59

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:16.748461+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:52:50.247480+00:00, confidence not recorded.
  - readme: https://github.com/hexo-ai/sia (fetched 2026-08-28T04:06:16.748461+00:00, sha 1e68c7ca4676)
  - homepage: https://hexolabs.com/ (fetched 2026-08-29T10:32:46.360100+00:00, sha 0ad1be7893cf)
- Data as of 2026-08-30T08:39:29.467469+00:00.