Write out which input contexts are loaded in the prompt file

Created on 15 November 2024, about 1 month ago

Problem/Motivation

Currently there is a no way to get a clear idea of what context is given into a prompt unless you are a developer that can follow code or if you have access to the logs.

Looking at the prompt and not seeing the this at first glance can be confusing since you will have no idea that we actually feed all field types installed into the field type selection prompt and you might assume its using its own knowledge/hallucinating/suggesting uninstalled fields.

Putting this in the prompt of service name and function name and placeholder variables would make it not just visible to that its being fed into the system, it also makes it so the function producing the prompts can fetch this which makes it possible to build the prompts in some developer tool without running the whole agent.

This will also take it closer to how OpenAI does Swarm with having the advantages of so called API Assistants Agents.

Proposed resolution

Add the possibilty to feed variables into the prompt creator function.
Add the possibility to declare service and method for each input context in the YAML.
Add the possibility to declare which variables are required in the YAML.
Make the prompt creator function run the context id if its missing.

Remaining tasks

User interface changes

Changes on how the prompts works, but you will still be able to dynamically feed prompts. Extra context is however needed into the function, since you might have variables needed for the methods.

API changes

Data model changes

Feature request
Status

Active

Version

1.0

Component

Code

Created by

🇩🇪Germany marcus_johansson

Live updates comments and jobs are added and updated live.
Sign in to follow issues

Merge Requests

Comments & Activities

Production build 0.71.5 2024