promptdesk / promptdesk Goto Github PK

Allow user to export logs from the repo. This includes RAW POST/GET requests to LLM APIs, formatted input, output, and variables included.

Reasoning: Developers should never be tied into a single platform or provider for AI-related data.

Files required:

• backend/src/routes/api/logs.ts (add backend download function)
• frontend/src/stores/LogStore.tsx (add front-end download function)
• frontend/pages/logs.tsx (add download button)

• Change color to red, to depict it's dangerous action
• ask for confirmation before performing action, as delete can't be reverted and if user clicks on delete by mistake he loses the data.

Implement a source code editor instead of using a HTML contenteditable DIV for the model editor.

Reasoning: Validating, writing, and reading JSON is challenging with no styling. A source code editor can help with this. In addition, it can be used for future source-code related features - like code execution, source code modification, etc.

Files:

• frontend/pages/models.tsx (modify the contenteditable pre tag)
• NEW CodeEditor.tsx file

Here is an example of a good code editor: https://microsoft.github.io/monaco-editor/

Query the database for identical prompt before executing a model.

Possible ideas:

• hash prompt and add hash to logs for easy retrieval
• use a cache:true attribute on the prompt-desk api call
• possibly log that a cached request was serverd?

handlebars.js is enabled for each prompt - create documentation for this

Make the input fields - input and dropdown similar styles - height and padding.

When we hover over the prompt name, it's not apparent that the input is disabled.

We shall have disabled (not allowed) cursor on disabled input.

Other pages have a line to divide heading/navbar and content

Logs page doesn't have that

• when we click on the label we the correct input field should be focused.

Watch video for reference -

Allow developers to add test data with variables.

Idea: Test data can be a JSON object with attributes as variable names. These attributes can map to to prompt variables. This would allow engineers to quickly test a dataset to see how well a certain prompt/model is performing.

This may require the following:

• a new database table for datasets
• a new REST API to retrieve, edit, and delete a dataset and their items.
• a new page on the front-end that lists datasets available.
• the option to use a dataset for a prompt

This issue will enhance the modals on our web app.

• Add close icon on top.
• When we press escape, it shall close the modal