pi vs SWE-agent
AI coding agents for customization and software engineering research
pi is a TypeScript toolkit and coding-agent CLI for using or building customizable AI agents. SWE-agent is a Python research tool that gives language models tools to attempt fixes for GitHub issues; its maintainers recommend mini-SWE-agent for new use.
pi: Build and Customize AI Coding Agents
Pi is a TypeScript toolkit for building and using AI agents, with a coding-agent CLI, terminal UI, and unified API for multiple LLM providers. Its extensible design suits developers who want to adapt an agent to their workflows.
SWE-agent: Use Language Models to Fix GitHub Issues
SWE-agent gives a language model tools to work on real software repositories, including diagnosing and attempting to fix GitHub issues. It is designed for software engineering research and configurable coding tasks, though its maintainers now recommend mini-SWE-agent for new use.
| pi | SWE-agent | |
|---|---|---|
| Language | TypeScript | Python |
| License | MIT | MIT |
| Stars | 112k | 20.5k |
| Forks | 14.2k | 2.2k |
| Last analyzed | Oct 3, 2026 | Oct 3, 2026 |
Key differences
- pi combines a multi-provider LLM API, agent runtime, terminal UI, and TypeScript SDK; SWE-agent focuses on using a model to inspect and modify repositories for issue-fixing tasks.
- pi is designed for workflow customization through extensions, skills, prompts, themes, and packages; SWE-agent offers YAML configuration and is designed to be adapted for research and custom tasks.
- pi supports scripting and integration through print and JSON modes, RPC, and an SDK; SWE-agent documents SWE-bench benchmarking and an EnIGMA mode for cybersecurity challenges.
- pi is written in TypeScript and requires Node.js 22.19 or newer; SWE-agent is written in Python.
- Both use the MIT license. pi has 112k stars and 14.2k forks, while SWE-agent has 20.5k stars and 2.2k forks.
- pi runs with its launching user's permissions and advises sandboxing for stronger isolation; SWE-agent cautions that generated changes need review and that outcomes depend on the model and configuration.
Choose pi if you…
- want a customizable coding-agent CLI and terminal interface for your workflow.
- need a TypeScript SDK or API to build agent applications and integrations.
- plan to automate agent runs with print or JSON output, or control them through RPC.
Choose SWE-agent if you…
- want to study how language models use repository tools to attempt GitHub issue fixes.
- need workflows for evaluating coding agents on repository issues or SWE-bench tasks.
- want to explore EnIGMA for cybersecurity challenges, using version 0.7 as advised.
This comparison is generated with AI from the OSRepos analyses of both projects. Always check each project's repository and documentation before choosing.