pi-extension-raptor-mini
pi extension that adds GitHub Copilot's Raptor Mini (oswe-vscode-prime) model to pi's github-copilot provider.
Package details
Install pi-extension-raptor-mini from npm and Pi will load the resources declared by the package manifest.
$ pi install npm:pi-extension-raptor-mini- Package
pi-extension-raptor-mini- Version
0.1.1- Published
- Apr 13, 2026
- Downloads
- 279/mo · 15/wk
- Author
- wseubring
- License
- MIT
- Types
- extension
- Size
- 8.1 KB
- Dependencies
- 0 dependencies · 2 peers
Pi manifest JSON
{
"extensions": [
"./index.ts"
]
}Security note
Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.
README
pi-extension-raptor-mini
Adds GitHub Copilot's Raptor mini model (oswe-vscode-prime) to pi.
Why
pi's model catalog is baked in at build time. Raptor mini went to public preview on 2025-11-10, after the catalog was last regenerated, so pi --list-models doesn't show it. This extension adds it at runtime.
When pi-ai picks it up upstream, this extension becomes a no-op (it detects the duplicate, warns, and exits).
Install
pi install npm:pi-extension-raptor-mini
The
npm:prefix is required.pi install pi-extension-raptor-miniis interpreted as a local filesystem path, not an npm package.
Alternatives:
pi install git:github.com/WSeubring/pi-extension-raptor-mini # from git
pi -e /path/to/pi-extension-raptor-mini # local checkout, no install
Use
pi --list-models | grep oswe
# github-copilot oswe-vscode-prime 264K 64K no no
pi --provider github-copilot --model oswe-vscode-prime -p "hello"
# or match by name substring:
pi --model raptor -p "hello"
You need to be logged into Copilot in pi (/login github-copilot inside the interactive UI, or pi prompts on first use). Any Copilot plan works — Free, Pro, Pro+, Business, Enterprise.
Model facts
| Field | Value | Source |
|---|---|---|
| Canonical id | oswe-vscode-prime |
zed#49514 |
| Name | Raptor mini (Preview) | GitHub Changelog 2025-11-10 |
| Base | Code-tuned GPT-5-mini | GitHub Changelog |
| Context | ~264,000 tokens | lobehub model-bank (GitHub does not publish officially) |
| Max output | ~64,000 tokens | Community estimate — no authoritative source |
| Tool calls | Supported | GitHub Changelog |
Heads-up
pi's github-copilot provider talks to Copilot's API with VS Code's own integration headers (Copilot-Integration-Id: vscode-chat, etc.). That's how Copilot models — including Raptor mini — are reachable at all. This behavior predates this extension and applies to every request pi's Copilot provider already makes; installing this extension doesn't change the request shape, just adds one model id to the catalog.
GitHub's AUP prohibits misrepresenting client identity; in practice, enforcement has targeted automation/volume abuse rather than client-identity alone. If you want zero ambiguity, use pi's anthropic / openai / google providers with your own API keys instead.