Imagine this:
There is a workflow with multiple agents/chatbots. Each chatbot works on a different instruction. I like to make many changes to those instructions. I would also love to monitor how outputs change. Is there a way to automatically save the instructions with the outputs to some kind of a database. Ideally, I’d like to have a database with every workflow action saved and in there each chatbot listed with its instruction and its output.
I know I could it do it manually but since we are automating… Anyone? Any ideas?
I’m thinking on a JSON file passed between the agents and in the meta part of the file the agents could write down their instructions and type, but passing so much data would eat my tokens too fast. Maybe there are other solutions to that. Are there?
In terms of a database you could use something like Airtable or Firestore.
Can you specify what you mean in terms of passing the instructions from one agent to another? Are they passing instructions from one agent to another performing a different task? Or are they passing instruction information from a previous version to the agent using the current version?
What is the purpose of passing the instructions and outputs? Is it so that the agents are taking note of what was done previously?
I want every prompt with every answer for all the chatbot agents being recorded automatically in a database.
This way I would be monitoring the workflow more efficiently.
Ah I see.
You can have the prompt be entered into a database along with it’s response when the automation runs. Since inputs cannot be directly mapped what you can do I set the prompt as a variable.
How do you plan to make the changes to your prompt?
I have all my prompts stored in a database and called upon when the scenario runs through certain filters. Changes to the prompts in this database will be reflected when the prompt is called for in a scrnario.
Thanks. The variable solution is something to think about.