Perplexity CLI

CLI for interfacing with Perplexity's API. Can also be used as a chatbot.

Prerequisites

• Perplexity account and API key. You'll be prompted for the API key the first time you run pplx.

Installation

With Homebrew

If you're using Homebrew:

brew install japelsin/tap/pplx

From source

If you have go installed:

go install github.com/japelsin/pplx@latest

Otherwise you could grab the appropriate executable from releases.

Usage

Search

The response is always streamed, all other parameters are available. The model is set through the config (see below).

Usage:
  pplx search [flags]

Flags:
  -f, --frequency_penalty int   How much to penalize token reuse. 1 is no penalty. Between 0 and 1.
  -m, --max_tokens int          Maximum number of tokens to be used per request. Defaults to config value. (default 1000)
  -p, --presence_penalty int    How much to penalize existing tokens. Between -2 and 2.
  -t, --temperature int         The amount of randomness in the response. Between 0 and 2.
  -K, --top_k int               Number of tokens to consider when generating tokens, lower values result in higher probability tokens being used. Between 0 and 2048.
  -P, --top_p int               Nucleus sampling. Probability cutoff for token selection, lower values result in higher probability tokens being used. Between 0 and 1.

Config

Configure pplx.

Usage:
pplx config [command]

Available Commands:
path        Get configuration file path
reset       Reset config
set         Set config value

Config values

• Additional instructions: Instructions appended to queries. If you intend to use pplx as a search engine it's recommended to instruct it to always provide its sources.

• Model: Available models are: pplx-7b-chat, pplx-70b-chat, pplx-7b-online, pplx-70b-online, llama-2-70b-chat, codellama-34b-instruct, codellama-70b-instruct, mistral-7b-instruct, and mixtral-8x7b-instruct.

• API Key: Self explanatory