clicky-chats

README

Clicky Chats

Clicky Chats is an OpenAI API clone with an extension to chat with tools.

Components

There are two basic components: server and agent. The server handles the basic CRUD operations and the agent is responsible for the orchestration. These can be run by specifying the corresponding subcommand. One can also run the server and agent in the same process by running clicky-chats server --with-agents.

Development

The two components can be run simultaneously for development with the following:

export OPENAI_API_KEY=<your-api-key>
make run-dev

If you need to create assistants that use the retrieval tool, you will also need to run the knowledge-retrieval-api service. See the Complimentary Services section for more information. In that case, you need to export CLICKY_CHATS_KNOWLEDGE_RETRIEVAL_API_URL=http://localhost:8000 before starting clicky-chats.

Setting the CLICKY_CHATS_DEBUG environment variable to anything will turn on debug logging:

export CLICKY_CHATS_DEBUG=1
export OPENAI_API_KEY=<your-api-key>
make run-dev

Complimentary Services

Rubra UI

The Rubra UI is a simple web interface that can be used to interact with the server.

Repository: https://github.com/acorn-io/rubra-ui

Requirements

• NodeJS
• Yarn

Setup

1. Install dependencies:

   yarn
   

2. Start the server:

   export NUXT_API=http://localhost:8080/v1 # this points back to the clicky-chats server
   export NUXT_API_KEY=sk-foobar # required to be set for the frontend, but not used at the moment
   yarn dev
   

Knowledge Retrieval API

The knowledge retrieval API is a simple API backed by a Vector Database that allows you to augment assistant's answers with information retrieved from documents you add to it. This is a requirement to use the retrieval tool when creating a new assistant.

Repository:https://github.com/gptscript-ai/knowledge-retrieval-api

Requirements

• Python 3.10+
• Docker & Docker Compose (for running the pgvector database)

Setup

1. Setup a Python virtual environment and install dependencies:

   python3 -m venv venv
   source venv/bin/activate
   pip install -r requirements.txt
   

2. Start the server (hot-reloading code) and database:

   make run-dev
   

Note

You have to start clicky-chats with the following environment variable pointing to the knowledge-retrieval-api:

export CLICKY_CHATS_KNOWLEDGE_RETRIEVAL_API_URL=http://localhost:8000

Extending the OpenAI API

There are three separate APIs served here:

• /v1 servers a copy of the OpenAI API
• /v1/rubra serves an extended OpenAI API, that is, the same objects with additional fields added. All objects from OpenAI that aren't extended can are also served here.
• /v1/rubra/x serves our own APIs that are used in conjunction with the above.

To add extra fields to existing OpenAI APIs, the GetExtendedAPIs in the extendedapi package is used. The current example in that package is adding the gptscript_tools field to the OpenAI Assistant object. In order to do that, we must add that field to the CreateAssistantObject, ModifyAssistantObject and the AssistantObject. The generator will add any additional fields to the existing OpenAI objects for the extension API.

To add net-new APIs, paths and components are added to the pkg/generated/rubrax.yaml OpenAPI spec. This spec will be combined with the OpenAI API and extended specs to create one unified spec that is used to generate the types and server.