Add support for reusing generated logic from LLM nodes and python execution

[rossaai]

As LLMs are non-deterministic, there are instances where they may not function properly at one moment but spontaneously resolve the issue at another. This inconsistency can be problematic, especially when an LLM node generates an incredibly effective solution to a problem, but later fails to reproduce the same result.

To address this issue, it would be beneficial to have a feature that allows users to construct a node using the AnyNode and then reuse or copy and paste the generated logic for utilization in other workflows without the need to make additional calls to the LLM server.

This feature would be different from caching the response, as it focuses on reusing the generated logic in any workflow by simply copying it, similar to running Python code directly in a node.

Implementing this feature would provide several benefits:

1. Consistency: Users can ensure that the generated logic remains consistent and reliable across different workflows and executions.
2. Cost-effectiveness: By reusing the generated logic, users can avoid the need to make repeated requests to the LLM server, reducing the associated costs.
3. Efficiency: Reusing the generated logic saves time and resources, as users can directly incorporate the proven solution into their workflows without relying on the LLM to regenerate it.

To implement this feature, consider the following steps:

1. Introduce a mechanism to capture and store the generated logic from an AnyNode after a successful execution.
2. Provide an option to export the captured logic in a format that can be easily copied and pasted into other workflows or nodes.
3. Develop a way to import and integrate the copied logic into a new node or workflow, ensuring compatibility and seamless execution.
4. Document the process and provide clear instructions to users on how to utilize this feature effectively.

[lks-ai]

Thanks for the input! A couple of things you might have missed about current functionality:

• We now have the Function Registry. It is stored in output/anynode folder and it's a json file which holds a backup of all of your functions by prompt md5 hashed. This will also tie into the chromadb store you might have noticed in the same folder for example recall on working functions. The registry only holds functions that have executed properly.
• Functions go through an iterative process using the function it just created in memory and this is how
• Just implemented "multiple attempts" where it goes 2 extra tries to get the code right, and then throws the error if it can't get it.
• IMPORT: If you use the same prompt as any remembered functions in the node registry, it will automatically use that function instead. This means you can get a node working, then use it in another workflow, or wherever as long as the prompt is exactly the same. This also lets you bundle up the function_registry json files from output with your workflows to let others use your functionality without LLMs. Not only doe that save you cost, it makes your node free for those you send the workflow bundle to.
• Forgetting: This one allows you to make the node forget it's function if you feel you've reached a dead end in iterative coding process for the node. All you have to do is delete the prompt (or cut it out) and run the queue once, causing it to reset it's state, then paste your prompt back in and continue.

All of this is getting refined and is in a state of flux right now as we are super-early in this project (8 days into development)

Node Export is on the roadmap.. I'm currently making sure the design of the system for this is correct, and ties into the rest of the planned nodes. There are a lot of planned features and all of them come from users like yourself who care about the project and want to contribute, so thanks a lot! It turns out, you are on the same page as everyone else :)

I Have a whole output of the planned features I'm working on where people can give their opinion in the discord: https://discord.com/channels/1244702057714421881/1244940700747825182/1244990597442310234 ... it's just me coding this so I have to take the features and order/prioritize them, but I'm getting through the list!

You're right about documentation. I wish I had more time, but it's been a week since the start of this and most of my time has been devoted to coding and making a few videos. I will work on getting the documentation better and work on making more videos that are informative and clear.

[lks-ai]

Noted and closed.

[rossaai]

Is there a way to run specific Python code in a node? For example, the output of any node, copy and paste it into a Python Any Node to run that code or other code that I wrote up

I've been looking for a solution to run Python on a node for weeks.