GitHub Next
Free tierGitHub's research lab exploring the future of software development
Free·All audiences·Powered by GitHub Copilot / OpenAI·Open source
Key strengths
Cutting-edge AI research prototypes for software developmentAgentic workflows and automation for GitHub ActionsHuman-in-the-loop code migration and verificationLLM-powered codebase understanding and navigationExploration of AI safety and integrity filtering for agentic systems
Completely free
San Francisco, USA
Founded 2021
No ratings yet
Technical Resources & Developer Integration
GitHub Next prototypes are designed to be studied, forked, and extended. Here's how to engage technically:
- Agentic Workflows: An open-source set of GitHub Actions workflows powered by natural-language instructions. Clone the repository, configure your GitHub token and LLM provider credentials, and reference the provided workflow YAML files as templates for your own repo automation.
- Integrity Filtering: A middleware layer for agentic pipelines that classifies untrusted GitHub artifacts (issues, PRs, comments) before they reach an LLM agent. Integrate it as a pre-processing step in any GitHub Actions-based agentic loop.
- Canary (Harm Gate): A small, auditable classification gate that sits in front of agentic workflows. Can be inserted as a GitHub Actions step; outputs a classification verdict that downstream steps can gate on.
- Crane (Code Migration): A migration assistant that plans, executes, and verifies code migrations in small agentic steps. Operates on a GitHub repository; designed to be invoked via CLI or Actions with a migration spec as input.
- Agent Functions: A pattern/library for structuring LLM agent prompts as reusable subroutines — reducing prompt bloat and enabling modular agent composition.
Most projects are published under open-source licenses on GitHub at github.com/githubnext. Check individual repositories for setup instructions, required environment variables, and usage examples.
