# xingyaoww/code-act

Official Repo for ICML 2024 paper "Executable Code Actions Elicit Better LLM Agents" by Xingyao Wang, Yangyi Chen, Lifan Yuan, Yizhe Zhang, Yunzhu Li, Hao Peng, Heng Ji.

Repository: https://github.com/xingyaoww/code-act
Canonical: https://ross.abutalabs.com/products/code-act
Language: Python
License: MIT
License Family: permissive
Topics: llm, llm-agent, llm-finetuning, llm-framework
Last push: 2024-05-23T23:14:59+00:00

## Health v2 (maintenance only)
Score: 26/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 68
- inputs: {"age_days": 963, "days_push": 832, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1699, forks 146 (observed 2026-08-28T04:05:24.115226+00:00)

## What it is
CodeAct is a research framework and agent system that uses executable Python code as a unified action space for LLM agents, enabling multi-turn interaction with a code interpreter. It includes the CodeActInstruct instruction-tuning dataset and fine-tuned CodeActAgent models (e.g., Mistral-7b based) that can be served via Ollama, llama.cpp, or Kubernetes.

## Use cases
- build llm agents that execute python code as actions
- fine-tune an open-source llm for agent tasks
- run a self-hosted chat agent with code execution
- evaluate llm agents on tool-use benchmarks
- deploy an llm agent stack on kubernetes
- run a code-executing agent locally on a laptop

## When to choose
- you want agents that act via executable code rather than JSON or text tool calls
- you need an open-source, fine-tunable agent model with a Python interpreter sandbox
- you want to reproduce or extend the CodeAct research (M3ToolEval, CodeActInstruct)

## When to avoid
- you need a production-hardened agent framework with broad integrations
- you cannot safely sandbox arbitrary code execution
- you only need simple prompt-based tool calling without code execution

## Facets
- artifact type: framework
- maturity: active
- function: agent-framework, llm-inference, llm-training, rag
- domain: artificial-intelligence, large-language-models, machine-learning
- platform: python, self-hosted
- tags: code-act, llm-agent, executable-code-actions, instruction-tuning, python-interpreter, research-paper, icml-2024, ai-agents, docker, kubernetes

## Member repositories
- xingyaoww/code-act (main) score 26

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:24.115226+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 2026-08-30T03:37:48.592504+00:00, confidence not recorded.
  - readme: https://github.com/xingyaoww/code-act (fetched 2026-08-28T04:05:24.115226+00:00, sha 334fe71ada2e)
- Data as of 2026-08-30T08:39:29.467469+00:00.