This is a build tool for creating a catalog intended for consumption by large language model chatbots.
Make sure your have dart installed. Follow the instructions, in the link below.
After installation, you can install the tools with the following command
dart pub global activate ok_ziggy_tools
Creates the catalog
Usage: zigt create [arguments]
-h, --help Print this usage information.
-i, --input The input file that contains domain names of the OpenAPI Spec service
(defaults to "domains.json")
The build generates a number of files including error logs. You will use some of these when constructing the chatbot server.
Copies the catalog files to a target directory
Usage: zigt copy [arguments]
-h, --help Print this usage information.
-t, --targetDir (defaults to "data")
Once you copy the file into a target directory, they will be nicely structured for a chatbot server to use.
The input file is just a list of domains hosting manifests. The tool will look for an ai-plugin.json manifest at the domain and then build a catalog entry for it.
[
"ai.abcmouse.com",
"api.speak.com",
"chatwithpdf.sdan.io"
]You can find reference to a couple of hundred services here: https://github.com/sisbell/chatgpt-plugin-store
There are two types of information that a chatbot needs to learn about services to provide to a user
- List of services - this describes what's available and allows the chatbot to decide what service to use based on client interation
- Specifications - this describes each endpoint and how to use and store remote data.
This is the primary service catalog that is used to advertise services to a chatbot.
[
{
"serviceId": "ff57602e9b",
"name": "ABCmouse",
"description": "Provides fun and educational learning activities for children 2-8 years old."
},
{
"serviceId": "495bc6a9bb",
"name": "Penrose Analyst",
"description": "Search global news and research papers. Summarize Arxiv.org links. Ask me for the latest news!"
},
{
"serviceId": "40dca1ffbe",
"name": "Crypto Market News",
"description": "It's your go-to solution for real-time cryptocurrency price updates, market insights, and the latest news."
}
]For each service there will be a spec file generated. It consists of two parts. The first part is the OpenAPI spec where the endpoint paths and request structure is defined.
The second part is the EXTRA_INFORMATION_TO_ASSISTANT. This is the description_for_model field from the AI manifest.
openapi: "3.0.0"
info:
version: "1.0.0"
title: "OK Ziggy"
description: "Serves as a dynamic proxy, enabling services like news, weather, travel and games."
servers:
- url: "httos://ziggy.zapvine.com/api/v1"
paths:
/services:
get:
summary: Retrieve a list of available services
operationId: "getServices"
responses:
"200":
description: Successful operation
# Rest of spec
EXTRA_INFORMATION_TO_ASSISTANT
The AI assistant's name is Ziggy. Ziggy's main job..[rest of prompt instruction] The following files are configuration files that a chatbot server uses to provide services
This maps the serviceId to the URL of the api server.
{
"ff57602e9b": "ai.abcmouse.com/ws/ai/0.1/gpt",
"a3be2a6602": "api.speak.com",
"2d59d0bd16": "chatwithpdf.sdan.io",
"5742d3c9b4": "openai.jettel.de",
"7da6597eed": "plugin.askyourpdf.com"
}This maps the service id to the domain that hosts the manifest. This domain considered the unique id of the service.
{
"ff57602e9b": "ai.abcmouse.com",
"a3be2a6602": "api.speak.com",
"2d59d0bd16": "chatwithpdf.sdan.io"
}This maps a service path to a content-type.
{
"ff57602e9b/ChatPluginRecommendActivities": "application/json",
"a3be2a6602/translate": "application/json",
"a3be2a6602/explainPhrase": "application/json",
"a3be2a6602/explainTask": "application/json",
"2d59d0bd16/loadPdf": "application/json"
}This command initializes the data directory needed to run a server
zigt init