One apply_patch for every model on Pi: GPT, Claude, Gemini, DeepSeek, Kimi, GLM, Qwen, and others. The extension replaces edit/write, choosing the grammar or JSON function tool based on provider capability.
Repo: https://github.com/paulpham157/pi-apply-patch
Native edit fails when the exact match string is off, while apply_patch applies the same intent with context matching:
 |
 |
Requires Pi >=0.85.1 and Node >=22.
pi install npm:@paulpham157/apply-patchIf Pi is running, use /reload. The extension activates automatically and keeps working across model switches.
- Create, edit, delete, and move files with a single patch tool.
- Confines paths to
cwd; rejects symlinks, existing destinations on create/move, and overlapping path operations. - Validates the full patch before writing; no full rollback on filesystem write errors.
Read a file before editing it; on patch failure, re-read the relevant region and generate a new patch. Patch usage quality varies by model. bash still works and is not subject to this tool's path limits.
Read CONTRIBUTION.md, AGENTS.md, and run the required tests before opening a PR. See live testing with Pi and releasing.
Issues? Open one here with the Pi version, provider/model, and reproduction steps. Do not attach API keys.
Quick create: Bug report · Model/provider issue · Feature request.
Forked from code-yeongyu/pi-apply-patch: at the time, upstream had no updates for ~2 months and only supported gpt-* models, so this fork was created for personal use. MIT license.