@Leon van Zyl is very rare for me to comment videos in youtube, this must be the second or max the third time in my life. The work you have done and the simplicity you explain the process deserve compliments. Well done and thank you
Thank you for what you do. I eagerly await the notification for the new video. May I ask for a hint about the optimal flow? Can data be processed iteratively or in batches using loops in Flowise? For instance, when transferring numerous articles from one DB column, having LLM handle it, and then writing it to another column in the DB, can this be done sequentially in batches?
OOOOOOOh sheeeet! This changed the game I built a new SAAS using crewAI , which works out 100x easier and more effective than Autogen, but still its code its not so easy to quickly test drive flows / different models, EXTRA tools!!! This is FKN UNREAL! Thank so much for bringing this to the table! You da best, learnt so much from you right at the start with the early Flowise tutorials you put out, I then learnt to make app with bubble, now with a no-code "crewAI" I can make any AI app imaginable! I owe you one Leon!!!
Leon, I have a problem where the flow never ends, it loops through the all the agents then starts again, looping infinitely... Do you have any ideas on that?
Great to see this video. And I have been going through all your videos and I have created a few chatbots. Can you also make a video on creating Conversational Website chatbot using Web scraper(fire crawl).
Another great video, thank you Leon! Since Flowise uses LangGraph to orchestrate the multi-agents workflows, how can such workflows be extended with Conditional Edges and Loops?
Hi Leon, quick question. With Autogen Studio, I can build an agent team that uses a tool\skill that can authenticate to Azure via the Graph API with specific authentication parameters (Tenant ID, Secret, etc...), and other agents can then write bespoke code that can use that authentication to answer my query e.g. how many users do not have MFA or add users to a group where the manager is Bob Smith. My question is "Can this same dynamic query\coding be created using a common authentication via Graph API with Flowise?" So far my Flowise experience has been that I have to create a custom tool for every conceivable scenario, as opposed to letting agents create and refine code dynamically and incorporate the correct Graph API authentication parameters to achieve the goal.
The "Prompt Engineering Team" isn't in the marketplace. Do I need to add/install it via other means? I'm using the latest version 1.8.3 Other than that, I've been able to follow your tutorials flawlessly. They're succinct and awesome. Thanks Leon!
Hi Leon, I hope you are doing great!! I'm glad to see that these new technologies are evolving and enabling solutions for more complex use cases. I will suggest 3 enhancements to the use case you presented that might take it a step further. And of course I kindly ask you to bring some light to the point by sharing your thoughts. So, how would you add: 1. The ability to get the list of target e-mails from a spreadsheet or database. (and/or pass it as a list in the prompt) 2. The ability to hide the process output to the end user (great for us to see what's happening but not very relevant to the end user that only wants the task done 😁) 3. The ability to use an e-mail template to format the e-mail in a very specific (corporate) way. And here I'm thinking about generating e-mails as well as other kinds of documents that might require following a pre-defined format. I hope you find these ideas interesting, and that you are able to bring some suggestions/guidelines to the table on how to achieve them. Thank you very much for your great work!!! Ricardo
Hi Leon, great work. I noticed that there is Record Manager node for AgentFlows to connect to Pinecone. How would I connect a Postgres Record Manager to avoid data duplication?
nice video , just wanted to get some demo on paid membership , i mean content topic and business use case studies , can get assistance on tech issues etc?
excellent video leon, thanks!!🙏 I have a problem with the chat, I can upload my document correctly, but when the worker reads the file it cuts the information in half, I need it to read 60 texts and generate a summary, but it only just give me 30 back
Try increasing the k number on the vector store node. Typically using a vector store to summarise data is counter intuitive, but you might have more luck if you increase the number of docs returned. You could also consider to upload the entire doc without any chunking.
Great video boss! . Btw Can u do a video on flowise where a customer chatbot send an automatic emails to client regarding the conversation had or any important information
You're welcome! This is an interesting use-case. I am planning to create a video on using agentflows with memory in the future, and this might be a good use-case for the tutorial.
Thank you for your help. "Make" and "Flowise", that is, it does not send mail even though there is no error on both sides. I don't receive mail. What could be the reason? I need help!
Hey Leon, great video! Is there a way to parse the information that is coming out from retrieving the document say in CSV format from the supervisor's end?
You would basically have to create two separate flows, where an LLM chain calls the agentflow and then parses the information. Happy to create a video on it, just can't think of a use-case? Mind sharing yours?
@@leonvanzyl Oh that makes sense! Thanks. My use-case is a bit complex, but in short, I am trying to retrieve information from a medical guideline document, and turn the result into a downloadable table for the providers.
Hey Leon, will you be doing a tutorial on how to use exported Chatgpt history to create long term memory flowise bot? Thanks for the great content btw. :)
Great video as always sir, please continue bringing content on Agents. Is it possible yet to have a human in the loop? Like to approve/reject agent's work before proceeding?
@@leonvanzyl thanks, I will keep an eye on it, I think once Langchain/Flowise allow a human in the loop to steer the agent's work, it will be terrific and an easier decision for companies. Thanks.
Thank you for the tut. Does it have any advantages before a single agent with multiple tools? Like executing RAG via Chain, Retriever or Chatflow tool?
You're welcome! Plenty of advantages. You could assign different models to each worker based on their role. For instance, a coder agent could have model assigned to it that is brilliant at producing code, etc. I like splitting the roles and responsibilities across different agents as apposed to forcing everything into a single agent.
You can prompt the different agents to be expert XYZ like in a company you would have different divisions or specialised workers and not just one person doing everything
Hi Leon, Quick question.When you deploy the multi agent team to a website, it does not show the individual agents at work. Is this by design? When I run the same question in flowise I see the agents working but on a site, I only see the final result.
Apparently Flowise added the reasoning steps to the Web Embed in the latest release. Haven't tested it myself, but check out this X post. x.com/FlowiseAI/status/1805986577887224124
Yeah, that's quite annoying. I wanted to create a video where I use the new Claude 3.5 Sonnet model, but Anthropic has a super tight rate limit. OpenAI seems to be the best option at this stage?
@@leonvanzyl for openai you will need to add/use $50 to level up to tier 2 to get increased limits (from 30k to 450k tokens per minute). Sadly, this limit is per project api key and not entire project as it was previously. Yea claude is limits are worse even for general number of queries for paid subscribers.
@@leonvanzyl ok maybe the marketplace has better examples now, but you don't really previously one's like you just did, they were all very basic before
when I connect retriever tool to the Worker I get: "Aborted! Error: Aborted! at agentNode (/opt/render/project/src/packages/components/dist/nodes/multiagents/Worker/Worker.js:239:15) at process.processTicksAndRejections (node:internal/process/task_queues:95:5) at async RunnableLambda.workerNode [as func] (/opt/render/project/src/packages/components/dist/nodes/multiagents/Worker/Worker.js:106:53) at async /opt/render/project/src/node_modules/.pnpm/@langchain+core@0.1.63/node_modules/@langchain/core/dist/runnables/base.cjs:1337:34 as well as 400 Invalid 'tools[0].function.name': string does not match pattern. Expected a string that matches the pattern '^[a-zA-Z0-9_-]+$'. Error: 400 Invalid 'tools[0].function.name': string does not match pattern. Expected a string that matches the pattern '^[a-zA-Z0-9_-]+$'. at APIError.generate (/opt/render/project/src/node_modules/.pnpm/openai@4.51.0_encoding@0.1.13/node_modules/openai/error.js:45:20) at OpenAI.makeStatusError (/opt/render/project/src/node_modules/.pnpm/openai@4.51.0_encoding@0.1.13/node_modules/openai/core.js:275:33) at OpenAI.makeRequest (/opt/render/project/src/node_modules/.pnpm/openai@4.51.0_encoding@0.1.13/node_modules/openai/core.js:318:30) at process.processTicksAndRejections (node:internal/process/task_queues:95:5) at async /opt/render/project/src/node_modules/.pnpm/@langchain+openai@0.0.30_encoding@0.1.13/node_modules/@langchain/openai/dist/chat_models.cjs:760:29 at async RetryOperation._fn (/opt/render/project/src/node_modules/.pnpm/p-retry@4.6.2/node_modules/p-retry/index.js:50:12) any ideas? If i run the whole setting at a normal Conversational retrieval QA Chain it works...
What did you name the tool? I think your tool name contains some invalid characters. It needs to be all lowercase with no spaces. Use underscrores (_) instead of spaces.
@leonvanzyl It is possible to create one, however I don't want to show the entire internal conversation in the chat, I want this conversation to be hidden from the user, and he only sees the final message
Hey there! That's interesting. I don't think it's possible to hide the steps in the chat window. If you're using API, you can extract only the final output.
@Leon van Zyl is very rare for me to comment videos in youtube, this must be the second or max the third time in my life. The work you have done and the simplicity you explain the process deserve compliments. Well done and thank you
I am honored to receive one of your comments 💓.
Thank you for the amazing feedback!
Your presentatiom is so perfect that makes me feel really lucky! Very well done, Leon!!! Congratulations!!!
Wow, thank you!!
You've truly made my day! Your remarkable effort in putting it all together is appreciated. Thank you!
You're welcome 🤗
Just what i needed right now. Good timing :)
Hehe, you're welcome 🤗
This is Gold❤ Thank you, Leon!! Excellent as always! ❤👌🙏🥇
Thank you! 😊
Awesome video Leon!
Love the use of the prompt engineering team! So helpful - My workers are working overtime :D
Thank you for what you do. I eagerly await the notification for the new video. May I ask for a hint about the optimal flow? Can data be processed iteratively or in batches using loops in Flowise? For instance, when transferring numerous articles from one DB column, having LLM handle it, and then writing it to another column in the DB, can this be done sequentially in batches?
OOOOOOOh sheeeet! This changed the game I built a new SAAS using crewAI , which works out 100x easier and more effective than Autogen, but still its code its not so easy to quickly test drive flows / different models, EXTRA tools!!! This is FKN UNREAL! Thank so much for bringing this to the table! You da best, learnt so much from you right at the start with the early Flowise tutorials you put out, I then learnt to make app with bubble, now with a no-code "crewAI" I can make any AI app imaginable! I owe you one Leon!!!
Awesome feedback!! Thank you 🙏
Leon, I have a problem where the flow never ends, it loops through the all the agents then starts again, looping infinitely... Do you have any ideas on that?
Can you make a video where you connect to a MySql DB and ask in natural language questions and retrieve data from the DB?
If you could mention how to connect to a SQL Postgres DB I would appreciate it!!
Yes please!
Great to see this video. And I have been going through all your videos and I have created a few chatbots.
Can you also make a video on creating Conversational Website chatbot using Web scraper(fire crawl).
This is a brilliant tutorial - thanks
Another great video, thanks Leon!
Thank you 🙏
Really useful, as always, Thanks so much, Lyon!!
You're welcome 🤗
All perfect thanks for sharing ! more FLOWAISE AI Please!!!
Will do
Awesome video Leon. That prompt engineering was particularly nice, that is a weakness of mine.
video showing calls with n8n would great to keep it all self hosted and open
Flowise is also self hosted and free 😊.
I'll definitely create videos on n8n soon. For interest sake, where do you host your n8n instance?
@@leonvanzyl I use AWS , or Render if I want a quick easy deployment
@@leonvanzyl hetzner cloud with coolify - no overages
Lekker dat ek jou kanaal ontdek het. Goeie materiaal, netjies uiteengesit, aangenaam om te leer. Sien uit na jou volgende aflewering!
Dankie Drikus!
This rocks!
So useful. Thanks!
You're welcome 🤗
thanks for the video,
you are only gem to enlighten us on flowise.
You're welcome
Another great video, thank you Leon! Since Flowise uses LangGraph to orchestrate the multi-agents workflows, how can such workflows be extended with Conditional Edges and Loops?
I know that conditional edges are coming soon. Will definitely create videos as these features roll out 👍
great video - thank you for sharing - Claude 3.5 is now available :) - I will try to do some mix diffrent models agents:)
Sounds awesome!
You solved the problem that I had witk Gmail API and Make before. Thank you very much!
Haha, glad I could help
Hi Leon, quick question. With Autogen Studio, I can build an agent team that uses a tool\skill that can authenticate to Azure via the Graph API with specific authentication parameters (Tenant ID, Secret, etc...), and other agents can then write bespoke code that can use that authentication to answer my query e.g. how many users do not have MFA or add users to a group where the manager is Bob Smith. My question is "Can this same dynamic query\coding be created using a common authentication via Graph API with Flowise?" So far my Flowise experience has been that I have to create a custom tool for every conceivable scenario, as opposed to letting agents create and refine code dynamically and incorporate the correct Graph API authentication parameters to achieve the goal.
Another great job!
Thank you 😊
The "Prompt Engineering Team" isn't in the marketplace. Do I need to add/install it via other means? I'm using the latest version 1.8.3
Other than that, I've been able to follow your tutorials flawlessly. They're succinct and awesome. Thanks Leon!
You know what, I think you might need to follow the "developer setup" to see this. Check out my Getting Started video for instructions.
Excellent presentation, thank you.
Thank YOU
Which is the better choice for agent-based workflows - Flowise or VectorShift?
Hi Leon, I hope you are doing great!!
I'm glad to see that these new technologies are evolving and enabling solutions for more complex use cases.
I will suggest 3 enhancements to the use case you presented that might take it a step further. And of course I kindly ask you to bring some light to the point by sharing your thoughts.
So, how would you add:
1. The ability to get the list of target e-mails from a spreadsheet or database. (and/or pass it as a list in the prompt)
2. The ability to hide the process output to the end user (great for us to see what's happening but not very relevant to the end user that only wants the task done 😁)
3. The ability to use an e-mail template to format the e-mail in a very specific (corporate) way. And here I'm thinking about generating e-mails as well as other kinds of documents that might require following a pre-defined format.
I hope you find these ideas interesting, and that you are able to bring some suggestions/guidelines to the table on how to achieve them.
Thank you very much for your great work!!!
Ricardo
Thank you for the suggestions!
@@leonvanzyl My pleasure!!
Do you believe that any of these can be implemented with the currently existing features in Flowise/Make?
Hi Leon, great work. I noticed that there is Record Manager node for AgentFlows to connect to Pinecone. How would I connect a Postgres Record Manager to avoid data duplication?
I have a dedicated video on Record Manager with Postgress that might interest you 👍.
th-cam.com/video/sNk6-ISi7i4/w-d-xo.html
nice video , just wanted to get some demo on paid membership , i mean content topic and business use case studies , can get assistance on tech issues etc?
Made my day
You're welcome 🤗
excellent video leon, thanks!!🙏
I have a problem with the chat, I can upload my document correctly, but when the worker reads the file it cuts the information in half, I need it to read 60 texts and generate a summary, but it only just give me 30 back
Try increasing the k number on the vector store node.
Typically using a vector store to summarise data is counter intuitive, but you might have more luck if you increase the number of docs returned. You could also consider to upload the entire doc without any chunking.
@@leonvanzyl I will do that, thank you 🙏😊
Thanks for making this! Have you had much success with other models besides ChatGPT?
Yes, the Anthropic models are brilliant!
Great video!!!
How can I create a worker that access a database?
Thanks a lot!!
Great video boss! .
Btw Can u do a video on flowise where a customer chatbot send an automatic emails to client regarding the conversation had or any important information
You're welcome!
This is an interesting use-case. I am planning to create a video on using agentflows with memory in the future, and this might be a good use-case for the tutorial.
@@leonvanzyl Thank you , That sounds perfect. Can't wait for that video 😃👌
@@leonvanzyl I second it, it would be great to get a transcript emails of the chats sent to the costumer and client.
Thank you for your help. "Make" and "Flowise", that is, it does not send mail even though there is no error on both sides. I don't receive mail. What could be the reason? I need help!
Hey Leon, great video! Is there a way to parse the information that is coming out from retrieving the document say in CSV format from the supervisor's end?
You would basically have to create two separate flows, where an LLM chain calls the agentflow and then parses the information.
Happy to create a video on it, just can't think of a use-case? Mind sharing yours?
@@leonvanzyl Oh that makes sense! Thanks. My use-case is a bit complex, but in short, I am trying to retrieve information from a medical guideline document, and turn the result into a downloadable table for the providers.
Hey Leon, will you be doing a tutorial on how to use exported Chatgpt history to create long term memory flowise bot? Thanks for the great content btw. :)
I'll see what I can do 🙂
Need more videos on Agents
Will do
Great video as always sir, please continue bringing content on Agents. Is it possible yet to have a human in the loop? Like to approve/reject agent's work before proceeding?
Will do 👍.
Not possible to create human in the loop [yet].
@@leonvanzyl thanks, I will keep an eye on it, I think once Langchain/Flowise allow a human in the loop to steer the agent's work, it will be terrific and an easier decision for companies. Thanks.
Baie dankie! Great content!
Awesome, you came through!
You're welcome 😁
@@leonvanzyl btw I am seeing a "Message Stopped" error. Am on Render - any way to solve this?
Am seeing that it's something to do with Docker updates; do we have to update the Docker settings on Render?
Nice 😊
Hi Leon. Great video.
Can I use azure instead of openai?
Yes, you can use any model that supports function calling
How can we keep the human in the loop to review and steer the output of the agents at each step?
Thank you for the tut. Does it have any advantages before a single agent with multiple tools? Like executing RAG via Chain, Retriever or Chatflow tool?
You're welcome!
Plenty of advantages. You could assign different models to each worker based on their role. For instance, a coder agent could have model assigned to it that is brilliant at producing code, etc.
I like splitting the roles and responsibilities across different agents as apposed to forcing everything into a single agent.
You can prompt the different agents to be expert XYZ like in a company you would have different divisions or specialised workers and not just one person doing everything
@@leonvanzyl Thank you!
Thanks a lot !
Awsome!
Great video
Thank you
Hi Leon, Quick question.When you deploy the multi agent team to a website, it does not show the individual agents at work. Is this by design? When I run the same question in flowise I see the agents working but on a site, I only see the final result.
Apparently Flowise added the reasoning steps to the Web Embed in the latest release. Haven't tested it myself, but check out this X post.
x.com/FlowiseAI/status/1805986577887224124
@@leonvanzyl Thanks
No, it does not seem to display the steps at all.. perhaps a bug they working on
@@WayneBruton yeah, could very well be a bug.
Just tested myself and I also don't see the reasoning.
Will log a message with them.
@@leonvanzyl Thanks Leon
is there way to add delay between each agent? Im getting hit with token rate limits. Also, the runs do not show up on langsmith.
Yeah, that's quite annoying. I wanted to create a video where I use the new Claude 3.5 Sonnet model, but Anthropic has a super tight rate limit.
OpenAI seems to be the best option at this stage?
@@leonvanzyl for openai you will need to add/use $50 to level up to tier 2 to get increased limits (from 30k to 450k tokens per minute). Sadly, this limit is per project api key and not entire project as it was previously. Yea claude is limits are worse even for general number of queries for paid subscribers.
i wish like comfy ui you can just drop a file into empty canvas and it populates all these nodes
You can import an existing flow, if that's what you mean?
The marketplace is full of templates that you can simply copy into your own space.
@@leonvanzyl ok maybe the marketplace has better examples now, but you don't really previously one's like you just did, they were all very basic before
Ek sal graag kontak met jou wou maak Leon!
My email is in die about section van channel, asook n link na my agentskap in description.
Flowise or langgraph or crewai ?
Agentflow is LangGraph
Just gold
Thank you
when I connect retriever tool to the Worker I get: "Aborted!
Error: Aborted!
at agentNode (/opt/render/project/src/packages/components/dist/nodes/multiagents/Worker/Worker.js:239:15)
at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
at async RunnableLambda.workerNode [as func] (/opt/render/project/src/packages/components/dist/nodes/multiagents/Worker/Worker.js:106:53)
at async /opt/render/project/src/node_modules/.pnpm/@langchain+core@0.1.63/node_modules/@langchain/core/dist/runnables/base.cjs:1337:34
as well as
400 Invalid 'tools[0].function.name': string does not match pattern. Expected a string that matches the pattern '^[a-zA-Z0-9_-]+$'.
Error: 400 Invalid 'tools[0].function.name': string does not match pattern. Expected a string that matches the pattern '^[a-zA-Z0-9_-]+$'.
at APIError.generate (/opt/render/project/src/node_modules/.pnpm/openai@4.51.0_encoding@0.1.13/node_modules/openai/error.js:45:20)
at OpenAI.makeStatusError (/opt/render/project/src/node_modules/.pnpm/openai@4.51.0_encoding@0.1.13/node_modules/openai/core.js:275:33)
at OpenAI.makeRequest (/opt/render/project/src/node_modules/.pnpm/openai@4.51.0_encoding@0.1.13/node_modules/openai/core.js:318:30)
at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
at async /opt/render/project/src/node_modules/.pnpm/@langchain+openai@0.0.30_encoding@0.1.13/node_modules/@langchain/openai/dist/chat_models.cjs:760:29
at async RetryOperation._fn (/opt/render/project/src/node_modules/.pnpm/p-retry@4.6.2/node_modules/p-retry/index.js:50:12)
any ideas?
If i run the whole setting at a normal Conversational retrieval QA Chain it works...
first of all thanks for the amazing content :*
What did you name the tool?
I think your tool name contains some invalid characters.
It needs to be all lowercase with no spaces. Use underscrores (_) instead of spaces.
@@leonvanzyl yep spacing.... no spacing in the title 🙈 thaaaaaaaanks
@leonvanzyl It is possible to create one, however I don't want to show the entire internal conversation in the chat, I want this conversation to be hidden from the user, and he only sees the final message
Hey there!
That's interesting. I don't think it's possible to hide the steps in the chat window.
If you're using API, you can extract only the final output.