@kushagharahi/pi-llama-extensions

Pi extensions for llama.cpp router — auto model discovery and tokens/second display

Package details

← Back
extension

Install @kushagharahi/pi-llama-extensions from npm and Pi will load the resources declared by the package manifest.

npm report
$ pi install npm:@kushagharahi/pi-llama-extensions
Package
@kushagharahi/pi-llama-extensions
Version
0.1.0
Published
Apr 26, 2026
Downloads
161/mo · 161/wk
Author
kushagharahi
License
unknown
Types
extension
Size
107.7 KB
Dependencies
0 dependencies · 2 peers
Pi manifest JSON
{
  "extensions": [
    "./extensions"
  ],
  "image": "https://raw.githubusercontent.com/kushagharahi/pi-llama-extensions/refs/heads/main/screenshot.png"
}

Security note

Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.

README

Pi extensions for llama.cpp power users

Features

  • Configure models only in llama.cpp
    • Auto model discovery in router mode -- you no longer have to duplicate what's in llama.cpp's models.ini in pi's models.json
    • Takes the first model as the default on load
  • Performace metrics in the Pi TUI
    • Tokens/second display
    • Prompt Processing % display

tok/s and prompt processing %

Quick start

pi install npm:@kushagharahi/pi-llama-extensions

models.json config for auto model discovery in router mode

Ensure that you have the following config for auto model discovery from llama.cpp's router mode. The two pieces of important info are the provider key llama-cpp and the model array having one "id": "llama-cpp-discover".

models.json:

{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://127.0.0.1:8080",
      "api": "openai-completions",
      "apiKey": "local",
      "models": [
        { "id": "llama-cpp-discover" } 
      ]
    }
  }
}

The extension will then autofill things like model id, name, contextLength, maxTokens.

Debug

Set LLAMA_CPP_EXTENSION_DEBUG=1 to enable verbose logging. Each extension writes to its own file:

Extension Log file
Auto Model Discovery /tmp/llama-cpp-auto.log
TPS Display /tmp/llama-cpp-tps.log

TPS Display also writes progress events to /tmp/llama-cpp-tps-progress.log.