I don’t understand why so many create more and more useless tools instead of cooperative work on a single open source truly great project… shame Thanks for the amazing vid
i have been extracting the heart of these models ( the are open source) as some are basically clones of each other : but i have notice the methodologys doing this : I have found to colaberate on a project you need a chat history : so after generation the model needs to reload the session : All the models use some form of agent like agen gen or crew aiwhichin turn use langchain ! - ( so we can bybass all of these ) - there is a very good tutorial by the code your own ai chanelwhere he talks about how to create the agent set up from scratch after these things came out they must of copied him ) ... so : you can use a notebook ! ( reallly its that simple i could not explain it better! ) - each project generated would be done using a note book ! so you can comeback to that session again relaoding the notebook ... ( behind the scenes ) . enabling for editing of the project ; So you would just send the model to its work folder to retrive it by its ID: or project name: or: Just create a command which executes system comands and python code functions : Create a single tool !: this tool enables the model to execute a code . save a code file ( create a tool ) , load a tool from file , and maintain its own list of tools : so it can add its tools to the list or remove them andlist the current tools: open interpretor also uses this technique to maintain its toolkit and load and unload tools etc editing them as it goes along ! ( installing what it needs ) So to create a project you can tell it to create a new folder and create the project in that folder ... it will ! your chain would need to be the three agents (Planner, Worker and Refiner ) And Manager ! . the storm etc was also ok but these model were using typescript so i skipped them as they are easy agent groups easy to devlop , even with a mini set of tools the model can research and create an enriched sumary , as well as query wikipedia instead of searching google !! storing the refferences as it travels , capturing the mian points then returning to grab the funll content after and sumarize the response ( research chain) i was able to work work with free chatgpt andcreate some agents androles andtools for the roles and some group tasks: so i used these to create a mini newsroom team to reseach topics and write storys based on a collection of given general topics , so it has a motive for the mnewsroom and it can when activated go out and do research (slow) and comeback with a basic newspaper full of articles : someof these models use a specific stack ... but you can goe through the code and find the prompts they used (often bad english hence bad performance) and adjust the prompt for your code stack... so these videos are really valuable and give great insights in how to clone the knowledge ! ( i noticed claude always considers if this question needs an artifact, if so make it so in the future the artifact can be refferenced hence my ( tooll kit function) ( let the o=maintain its own tool kit , loading the kit or even alowing it to scan the disk for current tools it may need.
Still very early days for these agent frameworks…. No clear “winner” yet, so it is fine that we all experiment with different approaches. Eventually a clear winner and approach will emerge that gains massive support from the community. I think Devinka is the current Open Source leader btw
It's because we are living in the experimental era. AI technology is incomplete so far and we are guinea pigs. It'll take several years to get real ones.
Thanks for this! I had so many problems running this and I thought it was just user error. You should revisit this once they update it because I think the feature they showcase in the repo to have MicroAgent visually build a website from an image could be useful eventually! Dropped you a super thanks for saving me some time.
@@AICodeKingthx for reply, i like that you showcase all of them. But its pretty hard for a viewer to judge there differences and usecase. I get that microagent does create a plan + tests it, which sounds like the more advanced approach, but if aider wins thats good to know. Could you maybe answer/make a video were you dont showcase them, but instead explain the pros/cons of each against the other and which one is good for what specific usecase ?
Also the claude API is not free now I used with aider But I encounter a problem that it will not work I have to purchase a plan Also with the live server I have used the port It does not work please help me guide or can you suggest me any videos
You should do a video using Aider+Claude+Prompt chaining to make an app. I got the idea from @indydevdan channel. I had Aider+Claude build a Python script to: 1. Take in my app idea's prompt and send it to claude. 2. Ask 3 clarifying questions about the app. 3. Build a multi step prompt chain to break the requests into smaller parts and save them to txt files to feed to the LLM sequentially. 4. Return errors seen in the terminal to the LLM for fixing after each step. It works pretty well. I'm surprised by how much better LLMs write code when you only give them a small step to work on each time.
I don’t understand why so many create more and more useless tools instead of cooperative work on a single open source truly great project… shame
Thanks for the amazing vid
i have been extracting the heart of these models ( the are open source) as some are basically clones of each other :
but i have notice the methodologys doing this :
I have found to colaberate on a project you need a chat history : so after generation the model needs to reload the session :
All the models use some form of agent like agen gen or crew aiwhichin turn use langchain ! - ( so we can bybass all of these ) -
there is a very good tutorial by the code your own ai chanelwhere he talks about how to create the agent set up from scratch after these things came out they must of copied him ) ...
so : you can use a notebook ! ( reallly its that simple i could not explain it better! ) - each project generated would be done using a note book ! so you can comeback to that session again relaoding the notebook ... ( behind the scenes ) . enabling for editing of the project ; So you would just send the model to its work folder to retrive it by its ID: or project name:
or:
Just create a command which executes system comands and python code functions :
Create a single tool !: this tool enables the model to execute a code . save a code file ( create a tool ) , load a tool from file , and maintain its own list of tools : so it can add its tools to the list or remove them andlist the current tools: open interpretor also uses this technique to maintain its toolkit and load and unload tools etc editing them as it goes along ! ( installing what it needs )
So to create a project you can tell it to create a new folder and create the project in that folder ... it will ! your chain would need to be the three agents (Planner, Worker and Refiner ) And Manager ! .
the storm etc was also ok but these model were using typescript so i skipped them as they are easy agent groups easy to devlop , even with a mini set of tools the model can research and create an enriched sumary , as well as query wikipedia instead of searching google !! storing the refferences as it travels , capturing the mian points then returning to grab the funll content after and sumarize the response ( research chain)
i was able to work work with free chatgpt andcreate some agents androles andtools for the roles and some group tasks: so i used these to create a mini newsroom team to reseach topics and write storys based on a collection of given general topics , so it has a motive for the mnewsroom and it can when activated go out and do research (slow) and comeback with a basic newspaper full of articles :
someof these models use a specific stack ... but you can goe through the code and find the prompts they used (often bad english hence bad performance) and adjust the prompt for your code stack...
so these videos are really valuable and give great insights in how to clone the knowledge ! ( i noticed claude always considers if this question needs an artifact, if so make it so in the future the artifact can be refferenced hence my ( tooll kit function) ( let the o=maintain its own tool kit , loading the kit or even alowing it to scan the disk for current tools it may need.
Still very early days for these agent frameworks…. No clear “winner” yet, so it is fine that we all experiment with different approaches. Eventually a clear winner and approach will emerge that gains massive support from the community. I think Devinka is the current Open Source leader btw
It's because we are living in the experimental era. AI technology is incomplete so far and we are guinea pigs. It'll take several years to get real ones.
Thanks for this! I had so many problems running this and I thought it was just user error. You should revisit this once they update it because I think the feature they showcase in the repo to have MicroAgent visually build a website from an image could be useful eventually! Dropped you a super thanks for saving me some time.
If you shoot the King, you better not miss...
AICodeKing: I don’t use tests because I believe you only live once! 😂👌
so what now? whats currently the best in your opinion???
microagent
OR
aider + deepseek
OR
aider + claude
OR
claude engineer?
Aider+Claude best, Aider+Deepseek 2nd, Claude Engineer, Microagent
@@AICodeKingthx for reply, i like that you showcase all of them.
But its pretty hard for a viewer to judge there differences and usecase.
I get that microagent does create a plan + tests it, which sounds like the more advanced approach, but if aider wins thats good to know.
Could you maybe answer/make a video were you dont showcase them, but instead explain the pros/cons of each against the other and which one is good for what specific usecase ?
not work for python
Thanks
"....I directly ssh into the server and directly edit code there...." YOLO! Too funny! I literally LOLed! 😃
Thanks!
Thanks a lot for the support!
The DeepSeekV2 version not working for me just hangs :(
Cool video! Please, create a video using SuperCoder from SuperAGI
Hi!
Do you know what's the best open source alternative to ParseLlama?
Thanks AI King.
Well, that saved me a lot of time thanks king. Currently, I'm trying to make a multi-modal ai assistant. Any suggestions?
Name it ReallyOpenAI (just kidding)
@AICodeKing I mean with the way it's working right now lol super hit or miss with the commands
Wrappers ! ( use that design pattern )
What about OpenDevin?
Haven't tested it in a while.
How can I view websites which are built in lightning ai vs code on html
Via custom port option. Use Live server VS Code Extension
Also the claude API is not free now I used with aider
But I encounter a problem that it will not work I have to purchase a plan
Also with the live server I have used the port
It does not work please help me guide or can you suggest me any videos
There is also ngrok port forwarding to a public URL but I don’t know if you want to do it that way. I would go by what AiCodeKing says
I cant run this using Deepseek LLM can i?
You can
😮😮
Nice to be here early
Something for uiux please 🥺❤️
E
Hey! Love your content - would love to sponsor a video. Can you let me know how I can contact you?
You can contact me on Discord : @aicodeking
Thanks
Thanks!
You should do a video using Aider+Claude+Prompt chaining to make an app. I got the idea from @indydevdan channel. I had Aider+Claude build a Python script to:
1. Take in my app idea's prompt and send it to claude.
2. Ask 3 clarifying questions about the app.
3. Build a multi step prompt chain to break the requests into smaller parts and save them to txt files to feed to the LLM sequentially.
4. Return errors seen in the terminal to the LLM for fixing after each step.
It works pretty well. I'm surprised by how much better LLMs write code when you only give them a small step to work on each time.
Great idea! BTW, Thanks for supporting
Thanks