Skip to main content

Comwork AI

Translations​

This tutorial is also available in the following languages:

Purpose​

This feature aims to expose AI1 models such as NLP2 or LLM3 to be exposed as an API based on this open source project.

Enabling this API​

In the SaaS version, you can ask to be granted using the support system.

If you're admin of the instance, you can grant users like this:

cwai_enable

UI chat​

Once you're enabled, you can try the CWAI api using this chat web UI:

cwai_chat

Use the API​

Of course, the main purpose is to be able to interact with those model using very simple http endpoints:

cwai_api

Here's how to get all the available models:

$ curl -X 'GET' 'https://cloud-api.comwork.io/v1/ai/models' -H 'accept: application/json' -H 'X-Auth-Token: XXXXXX'

Result:

{
  "models": [
    "gpt2",
    "nlptownsentiment",
    "mock"
  ],
  "status": "ok"
}

Then prompting with one of the available models:

curl -X 'POST' \
  'https://cloud-api.comwork.io/v1/ai/prompt' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -H 'X-Auth-Token: XXXXXX' \
  -d '{
  "model": "nlptownsentiment",
  "message": "This is bad !",
  "settings": {}
}'

The answer would be:

{
  "response": [
    "The predicted emotion is: Anger"
  ],
  "score": 1,
  "status": "ok"
}

Notes:

  • you have to replace the XXXXXX value with your own token generated with this procedure.
  • you can replace https://cloud-api.comwork.io by the API's instance URL you're using, with the CWAI_API_URL environment variable. For the Tunisian customers for example, it would be https://api.cwcloud.tn.

Use the CLI​

You can use the cwc CLI which provide a subcommand ai:

$ cwc ai
This command lets you call the CWAI endpoints

Usage:
  cwc ai
  cwc ai [command]

Available Commands:
  models      Get the available models
  prompt      Send a prompt

Flags:
  -h, --help   help for ai

Use "cwc ai [command] --help" for more information about a command.

List the available models​

$ cwc ai models
Models
[gpt2 nlptownsentiment nltksentiment textblobsentiment mock]

Send a prompt to an available model​

$ cwc ai prompt
Error: required flag(s) "message", "model" not set
Usage:
  cwc ai prompt [flags]

Flags:
  -h, --help             help for prompt
  -m, --message string   The message input
  -t, --model string     The chosen model
$ cwc ai prompt --model nltksentiment --message "This is bad"
Status  Response    Score
ok  [The predicted sentiment is: negative, score: -0.5423]  -0.5423

Driver interface​

If you fork the cwai API, you can implement you're own driver that will load and generate answer from models implementing this abstract:

class ModelDriver(ABC):
    @abstractmethod
    def load_model(self):
        pass

    @abstractmethod
    def generate_response(self, prompt: Prompt):
        pass

Then add your model in the ENABLED_MODELS environment variable.

In the future we'll propose a web GUI to upload your own drivers as a serverless function on the web console. In the meantime you can ask comwork via the support system.

  1. Artificial intelligence↩
  2. Natural language processing↩
  3. Large language model↩