apis

API definitions

File layout, running a test

For an api foo, the templates are:

• foo/lib/request-tpl.jsonnet and
• foo/lib/document-tpl.jsonnet.

There are also test files to check conversion.

cd foo
npm ci # once to setup the test project
npm test # run tests

Sample run

Using jsonnet command-line tool:

jsonnet --jpath ./lib ./openai/lib/request-tpl.jsonnet -V prompt=hello \
    -V secret1=TOKEN -V secret2=orgid

Output:

{
   "body": {
      "messages": [
         {
            "content": "hello",
            "role": "user"
         }
      ],
      "model": "gpt-3.5-turbo"
   },
   "headers": {
      "Authorization": "Bearer TOKEN",
      "Content-type": "application/json",
      "OpenAI-Organization": "orgid"
   },
   "method": "POST",
   "url": "https://api.openai.com/v1/chat/completions"
}

Request templates

Input:

• secret1: First secret, usually an API key
• secret2: Optionally second secret. In case of OpenAI API, it's an organization ID
• prompt
• parent: When using in a chain of templates, parent contains the output of the previous template. See openai-patch-example how to patch the transformation.

Output:

An object, ready to use for fetch. The expected fields:

• url
• method
• headers
• body: Type is the json object, not string

The header Content-type: application/json is often required by APIs, even if not mentioned in documentation.

There is a JSON schema for validation: ./schemas/request.json.

Document templates

Input:

• response: body payload

Output:

{
  doc: [
    {
      type: 'markdown',
      content: [
        {
          type: 'text',
          text: ... whatever ...
        }
      ]
    }
  ]
}

There could be several `type: markdown' objects.

There is a JSON schema for validation: ./schemas/document.json.

The document structure will be more rich in the future. It will be based on the ProseMirror's document model.

Using in NPM

npm install https://github.com/olpa/templating-for-api/releases/download/apis-v1.1.5/apis-v1.1.5.tar.gz