bytedance/SandboxFusion

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1060 stars · 103 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

A secure, self-hosted code sandbox service from ByteDance that runs and judges code generated by LLMs across 20+ programming languages via a uniform HTTP API. It ships with built-in support for popular coding evaluation datasets like HumanEval, MBPP, and CodeContests, and is designed for production cloud deployment with container-based security isolation.

Use cases

• run untrusted LLM-generated code in a secure sandbox
• evaluate LLM coding ability on HumanEval and MBPP benchmarks
• judge code outputs for reinforcement learning training
• execute code snippets in many languages through one HTTP API
• set up an online judge for coding evaluation datasets
• run GPU code like CUDA in an isolated environment
• serve a code execution backend for an AI coding agent

When to choose

• you need safe execution of model-generated code at scale
• you are benchmarking or training LLMs on coding tasks
• you want one API covering 20+ languages and standard evaluation datasets
• you can deploy Docker containers and want production-ready isolation

When to avoid

• you only need to run your own trusted code locally without isolation
• you cannot deploy Docker or privileged containers
• you need a lightweight in-process sandbox rather than a networked service

Facets

service · maturity active

testing benchmarking http-server security machine-learning llm-training gpu-computing large-language-models developer-tools testing artificial-intelligence self-hosted cloud-computing python self-hosted code-sandbox code-execution online-judge llm-evaluation code-judging multi-language humaneval code-contests jupyter-kernel rl-environment docker linux web-server gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="bytedance/SandboxFusion")