Skip to content

Latest commit

 

History

History
137 lines (109 loc) · 5.5 KB

README.md

File metadata and controls

137 lines (109 loc) · 5.5 KB
Raw

personal-ai-ts 中文

This is a personal ai tool that works with clients, such as Apple Shortcuts removing the need for a dedicated hardware device. it's begin from: fatwang2/siri-ultra, Sh4yy/personal-ai

Features

  • support search functions
  • support qianfan MaaS chat bot models

How it works

The assistant is run on Cloudflare Workers and can work with any LLM model.

Getting Started

  1. Clone the repository:

    • Clone this repository and navigate to the root directory.

  2. Install dependencies:

    • Run npm install to install the necessary dependencies.

  3. Authenticate with Cloudflare:

    • Run npx wrangler login to log in to your Cloudflare account.

  4. Create KV namespaces: (remote kv (expire_ttl) -> local/edge kv(cache_ttl))

    • Run npx wrangler kv:namespace create chats to create a KV namespace. Note down the ID, binding ai_chats.
    • Run npx wrangler kv:namespace create bots to create a KV namespace. Note down the ID, binding ai_bots.

  5. Configure the project:

    • Update wrangler.toml with the namespace IDs:
       [[kv_namespaces]]
   binding = "ai_chats"
   id = "<id>"
   [[kv_namespaces]]
   binding = "ai_bots"
   id = "<id>"

  6. Set up API keys:

  • When deploy to online, Run npx wrangler secret put API_KEY to set the Groq or OpenAI API key.

  • qianfan Maas app key:

    • run npx wrangler secret put QIANFAN_SECRET_KEY
    • run npx wrangler secret put QIANFAN_API_KEY

    Note:

    • You can only set API_KEY and set SEARCH_NAME is empty "" if you don't need search function
    • deploy online to set secret key, if dev, touch .dev.vars file, The .dev.vars file should be formatted like a dotenv file, such as KEY=VALUE
  1. Set up Functions API keys
  • Search, When deploy to online
    • Run npx wrangler secret put SERPER_API_KEY to set the serper API key.
    • Run npx wrangler secret put SEARCH_API_KEY to set the Search API key .
    • Run npx wrangler secret put SEARCH1_API_KEY to set the Search1API API key.
  • weather api key：
  1. Update the LLMs Vars: 
       [vars]
   API_BASE="https://api.groq.com/openai/v1/"
   MODEL="llama3-70b-8192"
   # output raw text prompt, need prompt tpl manager
   SYSTEM_PROMPT = "You are oligei. Answer in 1-2 sentences. Be friendly, helpful and concise. Default to metric units when possible. Keep the conversation short and sweet. You only answer in raw text, no markdown format. Don't include links or any other extras. Don't respond with computer code, for example don't return user longitude."
   #search_name: search_api | search1_api | serper_api
   SEARCH_NAME = "serper_api"
   #https://openweathermap.org/api/one-call-3#multi
   OPENWEATHERMAP_LANGUAGE = "en"
   #https://openweathermap.org/api/one-call-3#data
   OPENWEATHERMAP_UNITS = "metric"

    Tips: You can use ollama deploy local llm openai chat completions api

Deploying the Worker

To deploy the worker, run npx wrangler deploy.

NOTE

use other like https://personal-ai-ts.<your-username>.workers.dev/ api to set config api_key is not safety, just for local server. （if some app client want to add api_key, u need be careful）

Clients

Setting Up the Apple Shortcut

  1. Install the shortcut:

    • Use this link to install the shortcut, then edit it,

  2. Configure the shortcut: more detail: https://support.apple.com/zh-cn/guide/shortcuts/welcome/ios

    • Open the shortcut and replace the URL field with your worker's URL. If you didn't change the default name, the URL should be https://personal-ai-ts.<your-username>.workers.dev.
    • change URL field config param, see src/index.ts post / router api request params.

API

  • POST /: AI chat API.
  {
    "config": {
        "chat": {
            "api_base": "https://api.groq.com/openai/v1/",
            "model": "llama3-70b-8192",
            "system_prompt": "You are oligei. Answer in 1-2 sentences. Be friendly, helpful and concise. Default to metric units when possible. Keep the conversation short and sweet. You only answer in raw text, no markdown format. Don't include links or any other extras. Don't respond with computer code, for example don't return user longitude.",
            "api_key": ""
        },
        "qianfan": {
            "model": "completions",
            "api_key": "",
            "secret_key": ""
        },
        "search": {
            "search_name": "search_api",
            "search_api_key": "",
            "serper_api_key": "",
            "search1_api_key": ""
        },
        "weather": {
            "weather_name": "openweathermap_api",
            "openweahtermap_api_key": ""
        }
    },
    "location": {
        "longitude": 116.40,
        "latitude": 39.90
    },
    "chat_type": "chat_with_functions",
    "chat_bot": "openai",
    "chat_id": "349512281",
    "input": "today weather",
    "date": "2024/05/24 09:51"
}

tips: in local dev, use .dev.vars file define api_key from env,