I am planning a series Sequential Agents videos. Let me know which use-cases I should cover 👋. Remember to Like the video to support my channel 🙏. If you run into any issues when installing Flowise with npx flowise start, then try "npm install -g flowise" first. So the sequence would be: npm install -g flowise npx flowise start
I really love your video! I would like to see a workflow where an agent conducts an interview (e.g. with a job applicant) based on a job position. He keeps asking question until he has enough information to either accept or reject the applicant. He should then generate a file in JSON format to explain the result and the reasoning
could you please build app that show intermediate steps like multi-agents conversation, I think it will be so cool to see the discussion among multi-agent so that we can explain how we get the results (explainable AI) 🙏🙏🙏🙏🙏
Very interesting, thank you. I would be interested to know if these AI agents can actually create something like a quote, if the agents have access to the required information, e.g. a price list, retrieve the required items and create a Word or Excel file, and store it on my Computer. 😊
I've waited so long for this functionality. Thanks for the great tutorial. It really helped me to understand the workflow. I'm really looking forward to test it by myself. Once again, great job Leon. 🤩🙏
I know it’s really helpful to Ask AI, then tell it why its answer is not totally true, then have it find a balance between the answers to get a more critical output. I’d like to see reiterative processes so similarly/different prompted agents can provide eachother feedback before passing it down to the agent in the next step.
Well done Leon definitely going to play with the to play with these new sequential features. Thanks for your efforts. I like the assistants namen John 💪
Amazing series, you are my go-to for catching up with tools like Flowise! Question for you and/or everyone else tuning in and following along. I've built out the flow up to about the 20 min mark where we ask the name collector to enter a blank value in the additional parameters if someone enters "hi" or "hello". Unfortunately my flow continues to update the state with whatever the user enters ("hello", for instance) as the name regardless of these instructions. I'm running GPT-4o via azure api. Has anyone else had trouble with this step?
O unico problema desse video é que ele acaba..., por mim ficaria horas consumindo esse conteudo. Parabens você é bom, se tiver algum curso me mande, pois eu quero, sua didatica de ensino é muito boa, apesar de nao entender ingles mais com tradutor tudo se da um jeito !
Awesome! I just missed going deeper into condition agents (the second application) so that each agent has specific tools and documents to retrieve info from, because otherwise they make up every answer. Hopefully in next tutorials, thanks!!
Thanks, Leon! This is very helpful. I was wondering if it’s possible to integrate human interaction into the process, where we act as a guide or teacher and provide feedback at each step? For example, if I wanted to create a book with 5 different AI agents handling each phase, the process could look like this: 1. Idea Generation and Documentation: The AI (LLM) proposes ideas, and I give feedback for revisions until I’m ready to approve and move to the next step. 2. Story Definition: The AI suggests a story outline, and I provide feedback until I’m satisfied and give the go-ahead for the next phase. 3. Character Development: The AI creates character profiles, which I review and adjust until they meet my expectations, then I approve. 4. Chapter Outline: The AI proposes an outline for each chapter, which I review and refine until it’s ready to proceed. 5. Chapter Writing: The AI drafts each chapter, and I review and approve each one before moving to the next chapter. How would you structure this to ensure smooth human interaction at each step? I'm struggling to find a clear solution.
Your videos are great man, great that I never built using a langchain with coding but Flowise has some issues with local and production-grade. I can make more videos on langflow for production-grade applications
I've personally had issues with Langflow in prod.. which is kinda why I stopped creating videos on it at this stage. Whereas my FW instances has been running in prod for nearly 2 years without fail 😅. I'm curious, where you able to self host Langflow (the latest version)? I'd like to hear your thoughts on its memory requirements.
I would love to see a use case of sequential agents employed in a domain-driven design. An eCommerce example. The agent/s should have access to documentation which informs the domain model design. The documentation could be a picture of a business model canvas or access to detailed documentation such as Google docs etc.
Outstanding. It would be really nice to have a video that shows the user how to master HITL by using the LLM Node + 2 x Tool Node sequence. In the video (example), you could have a user ask a booking assistant to book a flight to London. The LLM Node, via a Tool Node, could automatically search for a flight (tool 1), and then a second Tool Node (with HITL enabled) could be used to execute the payment (tool 2), requiring permission from the user. Agent Nodes cannot do that since all the attached tools are subject to the same HITL setting, either enabled or disabled.
Now we can really build flows and combine the best of what the LLM can do with what older tools such as DialogFlow or VoiceFlow provided. The best of both worlds, because until now, neither of the two techniques was strong enough to control dialogs on its own.
I have some questions that weren't adressed in this video: - How do I get the Agent memory to work if I don't know SQL? - I can't find the startnode anywhere nor the 'agent'. Just variants of agents Starting the chat leaves me with: Cannot read properties of undefined (reading 'bind') TypeError: Cannot read properties of undefined (reading 'bind')
Thanks Leon. You are making great videos, clear and useful. I followed this process to make these agents using OpenAI and it worked. Then I switched out OpenAi and replaced it with the Anthropic module but it no longer worked. I have searched high and low for a solution but do not understand why they are not interchangeable. Do you know how Anthropic or Llama or Groc can be used instead of OpenAI to build sequential agents in Flowise?
@@leonvanzyl I actually ran into problems with OpenAI overcharging massively. What alternative LLM would you recommend for Agents to replace OpenAI. I tried Anthropic but it was not working with the setup you demonstrate in this video. Thanks for your advice!
This is very interesting. I will have to modify some applications to test and use this feature. I have some service manuals with diagrams. Is it possible to create a system that displays a specific diagram or refers the user to a specific URL?
Thank you for the video, I really appreciate it! Is there a way to dynamically update the stats table so it pulls the information from each field in Airtable and inserts it at the start of the conversation?
Great video! I have a question: If I create a chatbot in a local environment as you taught us, how can I configure it to work on a website? Specifically, how can I deploy it so that clients or visitors on the website can access the chatbot?
Both platforms offer unique features, so I would suggest checking out both. Both services offer a cloud solution now. If you wanted to self host, I would recommend going with Flowise. Langflow is a nightmare to self host at this stage.
Hi Leon, you're doing an amazing job. Great video as usual. How can I stop the agent from being repetitive when a question is asked for the first time? For instance when you setup the flow and asked "Hello", the agent's response was repetitive: "Hi there! How can I assist you today?" and "Hi there! How can I assist you today?". In fact I see that the response is always repetitive when the conversation is continued.
Did you assign the memory node by any chance? I would expect the agent to simply continue the conversation when the memory node is connected. I'll test this as well 👍
So very nice! Congrats man, excellent video! My only concern right now is with latency. In my tests, some messages, even a simple "hello" to the first agent took around 10 seconds. Does it depend on hardware only (cuz my machine is kinda slow) or is it about the model itself (i'm using gpto4-mini)? Seems like other types of flows ain't as slow as these like chatflows for instance
Thank you Victor. The model plays a big role in terms of performance. You should check out my Groq video where I compare the performance of Groq to GPT 4o Mini. I think it feels long because the responses are not streamed back.
@@leonvanzyl Then, the way the teacher's feedback is delivered to the write agent is, The comment is not delivered directly to the write agent, but is delivered in the form of being stored in the conversation memory? I don't know for sure, but in order for the teacher's feedback to be delivered after writing, isn't there a need for an instruction to recall the corresponding content in the system prompt of the write agent?
Hey Leon, thanks for the great vids. Question for you. What if I wanted to capture a few parameters (ie first, last, email, phone) and save them to state, could a single agent do this? Or would you recommend an agent per parameter ? If single agent, would I need a condition node and loop back to the agent? Or could the agent ask multiple questions until all parameters are extracted and filled?
Hey Leon, first of all i want to thank you for your videos. They brought me from absolute zero to actually thinking about solutions. About the example, where you use the chatflow to ask the users name i have one thing that keeps me puzzling. My idea is to expand the flow and get even more information about the user. The thing is that the flow gets stuck in the next condition, because the condition note sees that the name is not empty and routes it back to the same agent. It clicked for me why that happens because the node sends it to the first condition which is relevant. My Solution would be to implement some sort of priority in the condition note. So that if two or more conditions are present, you can set a priority where to route next. I hope it makes sense what i'm talking about. Is there a workaround for my idea or would it be possible to add that feature? Thanks again and best regards. Oliver
Hi Eule, Thank you for the wonder message. Glad to be a part of your learning journey. Please give me some time to look over your message. I fully understand your example, but it's a bit complicated to try and go over in a comment. I'll try to work the answer into a future video. You've given me a lot to chew on ☺️
Thank you! I have a Langflow series on my channel as well. Langflow is not quite as mature as Flowise, but I am covering their features as they roll out and stabilize.
Thank you for these great videos. I tried following this tutorial. When I add the LLM node and there is no "name", it goes to the "name collection agent" but this agent does not respond anything and no further message is obtained in the chat window at this state. If I remove the LLM node, it works fine as per the condition specified. What I could have missed here?
Hallo Leon. Ek love jou videos en leer regtig baie by jou. So dankbaar wat jy doen. Ek wil baie graag my idees in produksie sit maar ek weet glad nie wat die kostes gaan wees nie. Kan jy dalk 'n video maak oor wat mens alles nodig het en wat dit moontlik gaan kos om 'n besigheid te maak uit Flowise and AI apps? Sal jy dit dalk oorweeg om 'n vinnige call met my te hê? Ek bly in Pretoria. 🙏🏻
Hi Niel, baie dankie vir die positiewe terugvoer. Ek sal aanbeveel dat jy Flowise self-host of dat jy hulle cloud hosting gebruik. Self host is bietjie goedkoper. Loer bietjie na hierdie video. Kos so $7 pm en jy kan heelwat flows daarop hardloop. th-cam.com/video/OMNC8MQKosU/w-d-xo.html
Great video, thanks for the tremendous value! Unfortunately, I keep getting an error message. Although I have started from the beginning several times :( Can you please tell me where my error could be? This message appears when I test the flow: Error buildAgentGraph - 400 Invalid 'messages[2].name': string does not match pattern. Expected a string that matches the pattern '^[a-zA-Z0-9_-]+$'. Thank you very much. Best regards, Dean
this video is gold, just one simple question: How do I chain two Condition Nodes? I’m getting this error… Error buildAgentGraph - Condition conditionalEdge already present for node if_the_name_or_type_provided
Thank you for the video. Is it possible to create: a first ConditionAgent1 which will determine two values 'value1' or 'value2' If value1, then redirect to agent1 If value2, then redirect to a ConditionAgent2 ConditionAgent2 which will determine three values (actions) one agent per action ?? Thx
i have been searching and struggling trying to figure out how to make my chatflow "write file" node work. i want to take the output of the chatbot and save it to a file somehow but i cant find anything about the write file node literally anywhere. how do i interact with it?
Hi Leon love your stuff! Also, is there a way for the loop to run a tool every time? e.g. If loop one runs to ask who is Tesla's CEO and loop 2 does a 'verification' by running a bravesearch or google CSE search? It keeps throwing an error for me.
Hey MG, thanks for the feedback! Is a loop necessary for your scenario? You could just chain multiple agents together for each of those tasks. For interest sake, what's the error?
Hallo Leo...using flowise and chromaDb for a RAG chatbot (via phone), how can i set the "name collection" dynamically to connect a specific and always changing CHROMADB data path? i wish i could get each phone user to be recognized and directed to the correct data path using the caller ID which is unique. Thanks
@leonvanzyl Thank you for the video. Is it possible to create: a first ConditionAgent1 which will determine two values 'value1' or 'value2' If value1, then redirect to agent1 If value2, then redirect to a ConditionAgent2 ConditionAgent2 which will determine three values (actions) one agent per action
Hi Leon, great video about Sequential Agents! I'd like to try it but I'm still on Flowise version 1.3.12 (I followed you tutorial from last year and did an update on my GitHub repo, did a redeploy an it deployed nicely with v1.3.12 . As I'm not into render or setting up such services I don't know how to upgrade to Flowise 2.x. Can you give a hint? Does that go through updating v 1.3.12 ? Or do I have to create a new GitHup repo for version 2.x and then somehow deploy it to Render?
Error buildAgentGraph - 400 {"error":{"message":"Failed to call a function. Please adjust your prompt. See 'failed_generation' for more details.","type":"invalid_request_error","code":"tool_use_failed","failed_generation":"MAINTENANCE"}}
Leon, memory agent for now only uses sqlite like you did? Did you use the reference from a sqlite database when installing? Here we are using Postgres but it appears that it does not yet support sequential agent memory.
Hello! If I integrate Flowise for example with the N8N and it is connected to Whatsapp where several users will ask questions, how does flowise manage the context? Is there a way to pass the phone number as a parameter to be the conversation indicator?
I would use the phone number (which n8n receives from WhatsApp) as the "Session ID". This way the chatbot will be able to store and recall the conversation history.
Your initial understanding is correct. The flow moves in one direction. Unless you add a loop to the flow, the logic will not "return" to a previous agent. When the user sends a follow-up question, the entire flow is triggered again.
Hello, I'm new to this, but after watching this great tutorial twice, I would suggest inserting a "condition" node right after the "start" node, similar to the "hotel" example with 3 agents. The condition to evaluate would likely be something like "detect if the user is disapproving the essay produced," with results "yes" or "not," and then ending the conversation or redirecting the flow to the "3 loop essay production" again. However, I think it might be necessary to consider a state variable first to determine if the essay loop has already been run at least once. Or something like that 😅 It certainly is a new way to develop conversational flows... super We're all new here. Furthermore, I think I have also seen that within a "chatflow," we can call other "chatflows." I suppose that with agents, it is probably possible to define some kind of "agent orchestration." Then there should be the "3 loop producer," hehehe, which would be called by another agent or by the "orchestrator" agent (I think this is a "supervisor" agent). Man, we all need to dedicate a few hours to play with this!! 😅
Sad that it does not work, after adding the llm node after the condition node, i don recieve an text from the node behind anymore. So my Ai does not ask me for my name anymore.
You might be missing a few steps from the sound of it. The LLM node by itself won't produce much in terms of a response. Did you follow the video all the way to the end?
@@leonvanzyl yes i did, 3 times alredy, step by step, before i add the llm, my agent ask me for my name after introduce the llm. It dont ask anymore, i just get the empty Response
@@leonvanzyl I've now found the problem. I tried to replicate your tutorial, with llama3.2 via ollama. I tried nearly everything, but the last thing I haven't tested was the possibility, that a locally hosted model, would be not able to do the simple tasks asked from it. So I paid a few bucks for the openaiAPI and tested with that... It works!! I'm a bit sad tough, I would have loved tho get that thing working fully locally...
I am planning a series Sequential Agents videos. Let me know which use-cases I should cover 👋.
Remember to Like the video to support my channel 🙏.
If you run into any issues when installing Flowise with npx flowise start, then try "npm install -g flowise" first.
So the sequence would be:
npm install -g flowise
npx flowise start
I really love your video! I would like to see a workflow where an agent conducts an interview (e.g. with a job applicant) based on a job position. He keeps asking question until he has enough information to either accept or reject the applicant. He should then generate a file in JSON format to explain the result and the reasoning
could you please build app that show intermediate steps like multi-agents conversation, I think it will be so cool to see the discussion among multi-agent so that we can explain how we get the results (explainable AI) 🙏🙏🙏🙏🙏
Very interesting, thank you. I would be interested to know if these AI agents can actually create something like a quote, if the agents have access to the required information, e.g. a price list, retrieve the required items and create a Word or Excel file, and store it on my Computer. 😊
can you do something that connects to Airtable to save some SQL data?
Hello Leon, wonderful tutorial, and definitely interested in more videos. Just a quick one. Is there a way for Flowise to output in PDF format
This is exactly what I was looking for. Thank you.
Thank you very much for the support. It really helps my channel.
ME TOO
Best teacher ❤️ for Flowise. Please never stop increase your knowledge about flowise and teaching us.
Sincerely your student ❤
Thank you, I will
Truly amazing how fast you put out this comprehensive video after the flowise update, thanks!
It was a crazy amount of work and research. Thank you for noticing 😊.
That's both intuitive and interesting. Thanks for putting it together.
Congratulations on the great work you do on your channel. I wish you the success you deserve
This is incredible! Thank you for the super 🙏.
Really glad you enjoyed my content
Thx Leon. Although I have problems with English, you are the source I understand best.
That's a massive compliment. Thank you!
I've waited so long for this functionality. Thanks for the great tutorial. It really helped me to understand the workflow. I'm really looking forward to test it by myself. Once again, great job Leon. 🤩🙏
Glad it helped!
Thanks a lot
You're welcome 🤗
Thanks
Thank you for the support 🙏
I rarely watch videos twice, but you covered so much ground here, I will probably watch 3X so that I can do at least one lab-style solution.
Thank you! It was a lot of hard work and I tried to cram as much information into the video as possible
Always the best tutorials on Flowise! Thank you, this is awesome!
You're very welcome!
No words for all your work you are doing. Thanks a lot
You're welcome 🤗. Just glad I could help.
I know it’s really helpful to Ask AI, then tell it why its answer is not totally true, then have it find a balance between the answers to get a more critical output. I’d like to see reiterative processes so similarly/different prompted agents can provide eachother feedback before passing it down to the agent in the next step.
Definitely. Agentic RAG is a good example of this. Tutorial coming soon!
Another great video from Leon. Greetings from Argentina.
Thank you!
Howzit, from South Africa
Another great video Leon!
Well done Leon definitely going to play with the to play with these new sequential features. Thanks for your efforts. I like the assistants namen John 💪
Amazing series, you are my go-to for catching up with tools like Flowise! Question for you and/or everyone else tuning in and following along. I've built out the flow up to about the 20 min mark where we ask the name collector to enter a blank value in the additional parameters if someone enters "hi" or "hello". Unfortunately my flow continues to update the state with whatever the user enters ("hello", for instance) as the name regardless of these instructions. I'm running GPT-4o via azure api. Has anyone else had trouble with this step?
Thanks Leon for the hardwork you have done. Just a quick one, how can you connect sequential agents to a Vector Store? Thanks!
Thank you very much for the support!
Check out this video
th-cam.com/video/SL77Ojbgy6U/w-d-xo.htmlsi=iZIB2_WzrEQsIq_z
@@leonvanzyl This is what I'm looking for. :-) Legend.
mind blown! 🔥🔥✌✌..so much potential power there...amazing video thanks - the actual langgraph code is very complex so no chance without FW for me
Thank you Mu
O unico problema desse video é que ele acaba..., por mim ficaria horas consumindo esse conteudo. Parabens você é bom, se tiver algum curso me mande, pois eu quero, sua didatica de ensino é muito boa, apesar de nao entender ingles mais com tradutor tudo se da um jeito !
Amazing feedback, thank you!!
É verdade!! Mas o inglês dele é muito bom de entender, comparando com outros youtubers. O Leon é fera! Valeu Dantas!
Too Good.. didn't realize that was more than 30 minutes and more.. Cheers
Tried to cram as much value into the video as possible. There are timestamps so you can come back and reference a specific node in the future ☺️
Amazing Leon, I really needed this tutorial. Thanks so much
Awesome! I just missed going deeper into condition agents (the second application) so that each agent has specific tools and documents to retrieve info from, because otherwise they make up every answer. Hopefully in next tutorials, thanks!!
My latest video uses conditional agents. You might be interested in that.
Yet another great video, Leon! Thanks :)
Glad you enjoyed it
I was waiting for this.. I knew it was on its way :)
Haha, this video was a lot of work. Hope it was worth the wait 😁
Excellent tutorial video! Thanks for your effort. Save me lots of time from reading Flowise's documentation.
Appreciate it! Glad I could help.
Hi Leon, Great tutorial. A bit like using crew ai but without writing all the code. Would love to see more tutorials on this.
Exactly!
Excellent, looking forward to seeing more examples! If possible, please demonstrate the improved rag example.
Will do!
Thanks, Leon! This is very helpful. I was wondering if it’s possible to integrate human interaction into the process, where we act as a guide or teacher and provide feedback at each step? For example, if I wanted to create a book with 5 different AI agents handling each phase, the process could look like this:
1. Idea Generation and Documentation: The AI (LLM) proposes ideas, and I give feedback for revisions until I’m ready to approve and move to the next step.
2. Story Definition: The AI suggests a story outline, and I provide feedback until I’m satisfied and give the go-ahead for the next phase.
3. Character Development: The AI creates character profiles, which I review and adjust until they meet my expectations, then I approve.
4. Chapter Outline: The AI proposes an outline for each chapter, which I review and refine until it’s ready to proceed.
5. Chapter Writing: The AI drafts each chapter, and I review and approve each one before moving to the next chapter.
How would you structure this to ensure smooth human interaction at each step? I'm struggling to find a clear solution.
Thanks!
Thank you very much for the support ❤️
Your videos are great man, great that I never built using a langchain with coding but Flowise has some issues with local and production-grade. I can make more videos on langflow for production-grade applications
I've personally had issues with Langflow in prod.. which is kinda why I stopped creating videos on it at this stage.
Whereas my FW instances has been running in prod for nearly 2 years without fail 😅.
I'm curious, where you able to self host Langflow (the latest version)? I'd like to hear your thoughts on its memory requirements.
Even two months later, this video is priceless
🙏
i need more videos about this
thanks for this great video
Amazing. Thanks Leon
I would love to see a use case of sequential agents employed in a domain-driven design. An eCommerce example. The agent/s should have access to documentation which informs the domain model design. The documentation could be a picture of a business model canvas or access to detailed documentation such as Google docs etc.
Great content, Leon! A question if I may - when would you use supervisor/workers and an agent loop? Have you figured out any simple rule?
Outstanding.
It would be really nice to have a video that shows the user how to master HITL by using the LLM Node + 2 x Tool Node sequence. In the video (example), you could have a user ask a booking assistant to book a flight to London. The LLM Node, via a Tool Node, could automatically search for a flight (tool 1), and then a second Tool Node (with HITL enabled) could be used to execute the payment (tool 2), requiring permission from the user.
Agent Nodes cannot do that since all the attached tools are subject to the same HITL setting, either enabled or disabled.
Excellent example. Thank you!
Now we can really build flows and combine the best of what the LLM can do with what older tools such as DialogFlow or VoiceFlow provided. The best of both worlds, because until now, neither of the two techniques was strong enough to control dialogs on its own.
Thank you so much! Extremely helpful
You're welcome!
Thanks Leon that very useful 👍
This is great thanks!
Great work. For the loop node, I wonder if the teacher could give it a score and then add a score to the conditional.
Definitely. The grade could be stored as a state value. The loop could then check the grade in state.
Really fantastic stuff!! Can you compare this functionality to any pf the other tools on the market?
Amazing content, I LOVE YOU!
I have some questions that weren't adressed in this video:
- How do I get the Agent memory to work if I don't know SQL?
- I can't find the startnode anywhere nor the 'agent'. Just variants of agents
Starting the chat leaves me with:
Cannot read properties of undefined (reading 'bind')
TypeError: Cannot read properties of undefined (reading 'bind')
Thanks Leon. You are making great videos, clear and useful. I followed this process to make these agents using OpenAI and it worked. Then I switched out OpenAi and replaced it with the Anthropic module but it no longer worked. I have searched high and low for a solution but do not understand why they are not interchangeable. Do you know how Anthropic or Llama or Groc can be used instead of OpenAI to build sequential agents in Flowise?
You're welcome 🤗.
Yeah, the OpenAI models really good at reasoning and perfect for agents.
@@leonvanzyl I actually ran into problems with OpenAI overcharging massively. What alternative LLM would you recommend for Agents to replace OpenAI. I tried Anthropic but it was not working with the setup you demonstrate in this video. Thanks for your advice!
Thanks Leon!!
This is very interesting. I will have to modify some applications to test and use this feature. I have some service manuals with diagrams. Is it possible to create a system that displays a specific diagram or refers the user to a specific URL?
Thank you very much Leon.
Thank you for the video, I really appreciate it! Is there a way to dynamically update the stats table so it pulls the information from each field in Airtable and inserts it at the start of the conversation?
Surely you can add a custom tool at the start of the flow to retrieve the airtable records and then append the results to state. Something like that 😊
Great video! I have a question: If I create a chatbot in a local environment as you taught us, how can I configure it to work on a website? Specifically, how can I deploy it so that clients or visitors on the website can access the chatbot?
Thank you!
You can use the paid cloud service or self host it.
Here's a video:
th-cam.com/video/OMNC8MQKosU/w-d-xo.html
What are possible factors to consider when trying to decide between using Flowise or Langflow?
Both platforms offer unique features, so I would suggest checking out both.
Both services offer a cloud solution now. If you wanted to self host, I would recommend going with Flowise. Langflow is a nightmare to self host at this stage.
@@leonvanzyl Thanks for the response. You are killing it with this niche.
Hi Leon, you're doing an amazing job. Great video as usual. How can I stop the agent from being repetitive when a question is asked for the first time? For instance when you setup the flow and asked "Hello", the agent's response was repetitive: "Hi there! How can I assist you today?" and "Hi there! How can I assist you today?". In fact I see that the response is always repetitive when the conversation is continued.
Did you assign the memory node by any chance? I would expect the agent to simply continue the conversation when the memory node is connected.
I'll test this as well 👍
@@leonvanzyl Yes I did. Thanks for looking into it.
thank you for the update:)
So very nice! Congrats man, excellent video! My only concern right now is with latency. In my tests, some messages, even a simple "hello" to the first agent took around 10 seconds. Does it depend on hardware only (cuz my machine is kinda slow) or is it about the model itself (i'm using gpto4-mini)? Seems like other types of flows ain't as slow as these like chatflows for instance
Thank you Victor.
The model plays a big role in terms of performance. You should check out my Groq video where I compare the performance of Groq to GPT 4o Mini.
I think it feels long because the responses are not streamed back.
I enjoyed the video. I have a question. Even when the loop is run like now, are the teacher's critique results reflected in the re-write?
Correct, the teachers feedback is incorporated into the rewrite
@@leonvanzyl Then, the way the teacher's feedback is delivered to the write agent is,
The comment is not delivered directly to the write agent, but is delivered in the form of being stored in the conversation memory?
I don't know for sure, but in order for the teacher's feedback to be delivered after writing, isn't there a need for an instruction to recall the corresponding content in the system prompt of the write agent?
Would love to see how the flowise ai python interpreter tool works
Hey Leon, thanks for the great vids. Question for you. What if I wanted to capture a few parameters (ie first, last, email, phone) and save them to state, could a single agent do this? Or would you recommend an agent per parameter ?
If single agent, would I need a condition node and loop back to the agent? Or could the agent ask multiple questions until all parameters are extracted and filled?
Leon that was another awesome video.
Can you point me to a recording if have already done where we can use mongodb for vector database
Thank you!
Don't have a mongoDB video yet. I'll see what I can do 😉
Hey Leon, first of all i want to thank you for your videos. They brought me from absolute zero to actually thinking about solutions.
About the example, where you use the chatflow to ask the users name i have one thing that keeps me puzzling. My idea is to expand the flow and get even more information about the user. The thing is that the flow gets stuck in the next condition, because the condition note sees that the name is not empty and routes it back to the same agent. It clicked for me why that happens because the node sends it to the first condition which is relevant. My Solution would be to implement some sort of priority in the condition note. So that if two or more conditions are present, you can set a priority where to route next. I hope it makes sense what i'm talking about. Is there a workaround for my idea or would it be possible to add that feature? Thanks again and best regards. Oliver
Hi Eule,
Thank you for the wonder message. Glad to be a part of your learning journey.
Please give me some time to look over your message.
I fully understand your example, but it's a bit complicated to try and go over in a comment. I'll try to work the answer into a future video. You've given me a lot to chew on ☺️
Super helpful
Excelente
Really nice and informative video. Why *did you choose flowise over langflow?
Thank you!
I have a Langflow series on my channel as well. Langflow is not quite as mature as Flowise, but I am covering their features as they roll out and stabilize.
Thank you for these great videos. I tried following this tutorial. When I add the LLM node and there is no "name", it goes to the "name collection agent" but this agent does not respond anything and no further message is obtained in the chat window at this state. If I remove the LLM node, it works fine as per the condition specified. What I could have missed here?
How do you add custom tools with sequential agents. The custom tool node doesn't seem to work with the agent node.
Hallo Leon.
Ek love jou videos en leer regtig baie by jou. So dankbaar wat jy doen.
Ek wil baie graag my idees in produksie sit maar ek weet glad nie wat die kostes gaan wees nie. Kan jy dalk 'n video maak oor wat mens alles nodig het en wat dit moontlik gaan kos om 'n besigheid te maak uit Flowise and AI apps? Sal jy dit dalk oorweeg om 'n vinnige call met my te hê? Ek bly in Pretoria. 🙏🏻
Hi Niel, baie dankie vir die positiewe terugvoer.
Ek sal aanbeveel dat jy Flowise self-host of dat jy hulle cloud hosting gebruik. Self host is bietjie goedkoper.
Loer bietjie na hierdie video. Kos so $7 pm en jy kan heelwat flows daarop hardloop.
th-cam.com/video/OMNC8MQKosU/w-d-xo.html
Cool leon! Could you also show how to use langraph in python?
Working on a LangGraph series actually 👍
@@leonvanzyl nice man! Keep up the good worl 💪
User auth should be a nice to have.
Please make tutorial to how guardrails can be done in flowise.
why did you use LLM nodes for the loop example rather than Agent nodes?
Hi, thanks! How to split up the user sessions with overrides using Agent Memory as we can do with Memory nodes?
Great video, thanks for the tremendous value! Unfortunately, I keep getting an error message. Although I have started from the beginning several times :(
Can you please tell me where my error could be?
This message appears when I test the flow:
Error buildAgentGraph - 400 Invalid 'messages[2].name': string does not match pattern. Expected a string that matches the pattern '^[a-zA-Z0-9_-]+$'.
Thank you very much. Best regards, Dean
Receiving: Error buildAgentGraph - TypeError: Cannot read properties of undefined (reading 'includes') after including Conditional Agent.
this video is gold, just one simple question: How do I chain two Condition Nodes? I’m getting this error… Error buildAgentGraph - Condition conditionalEdge already present for node if_the_name_or_type_provided
Add an LLM node in between the two condition nodes.
Fiz a assinatura de membro aonde estao os videos exclusivos ?
Can u tutorial connect mysql or postgeree?
Will do.
Many people want to see some sort of SQL loader tutorial.
@@leonvanzyl I hope you can make a tutorial from installation, connecting Postgree to reading SQL using natural language :D
Thank you for the video. Is it possible to create:
a first ConditionAgent1 which will determine two values 'value1' or 'value2'
If value1, then redirect to agent1
If value2, then redirect to a ConditionAgent2
ConditionAgent2 which will determine three values (actions)
one agent per action
??
Thx
Would you be able to export this flowise demo file to a github repo for our reference...
i have been searching and struggling trying to figure out how to make my chatflow "write file" node work. i want to take the output of the chatbot and save it to a file somehow but i cant find anything about the write file node literally anywhere. how do i interact with it?
Is there any reason to use multi agents now we have sequential?
Hi Leon love your stuff! Also, is there a way for the loop to run a tool every time? e.g. If loop one runs to ask who is Tesla's CEO and loop 2 does a 'verification' by running a bravesearch or google CSE search? It keeps throwing an error for me.
Hey MG, thanks for the feedback!
Is a loop necessary for your scenario? You could just chain multiple agents together for each of those tasks.
For interest sake, what's the error?
Sometimes we are facing "reading length" errors. Appreciate if you can explain this error additionally. Thank you .
I haven't seen that error myself. You should upload your flow in the Flowise Discord channel.
Could you please share an export of this (without your API keys) that way viewers can build off what you already did?
Hallo Leo...using flowise and chromaDb for a RAG chatbot (via phone), how can i set the "name collection" dynamically to connect a specific and always changing CHROMADB data path? i wish i could get each phone user to be recognized and directed to the correct data path using the caller ID which is unique.
Thanks
@leonvanzyl Thank you for the video. Is it possible to create:
a first ConditionAgent1 which will determine two values 'value1' or 'value2'
If value1, then redirect to agent1
If value2, then redirect to a ConditionAgent2
ConditionAgent2 which will determine three values (actions)
one agent per action
Hi Leon, do you know if the state node is connected to the session ID, so the variables can differ from each session?
That is my understanding as well.
@@leonvanzyl Thanks Leon! Makes sense.
hi leon, do you have something like for LangFlow?
Hi Leon, great video about Sequential Agents! I'd like to try it but I'm still on Flowise version 1.3.12 (I followed you tutorial from last year and did an update on my GitHub repo, did a redeploy an it deployed nicely with v1.3.12 .
As I'm not into render or setting up such services I don't know how to upgrade to Flowise 2.x. Can you give a hint? Does that go through updating v 1.3.12 ? Or do I have to create a new GitHup repo for version 2.x and then somehow deploy it to Render?
I have a dedicated upgrade guide video that might interest you.
th-cam.com/video/46DDuc9MNJk/w-d-xo.html
@@leonvanzyl many thanks. I watched it and your instructions worked super well!
This is basically Langraph, right?
Absolutely right. It uses LangGraph under the hood.
Error buildAgentGraph - 400 {"error":{"message":"Failed to call a function. Please adjust your prompt. See 'failed_generation' for more details.","type":"invalid_request_error","code":"tool_use_failed","failed_generation":"MAINTENANCE"}}
I don't know where i am mistaken. It's coming in everything. I check like 10 times.
Leon, memory agent for now only uses sqlite like you did? Did you use the reference from a sqlite database when installing?
Here we are using Postgres but it appears that it does not yet support sequential agent memory.
It's SQLite only at this stage.
I'll definitely create a video if they roll out support for more DBs.
Can I conect the Agent node with Rag?
Absolutely!!
Hey Leon, How can we enable our chatbot to read database data directly from external database?
There's been a big demand for me to create a video where the bots use a database as the source.
Will make an effort to create a video soon.
Search on the nodes list, you have there a "SQL Database chain" which let you connect directly to a local/remote SQL database.
Hello! If I integrate Flowise for example with the N8N and it is connected to Whatsapp where several users will ask questions, how does flowise manage the context? Is there a way to pass the phone number as a parameter to be the conversation indicator?
I would use the phone number (which n8n receives from WhatsApp) as the "Session ID". This way the chatbot will be able to store and recall the conversation history.
I thought that in the sequential agent once the flow goes to the next agent it does not return. But I see that this happens. How to avoid this?
Your initial understanding is correct. The flow moves in one direction. Unless you add a loop to the flow, the logic will not "return" to a previous agent.
When the user sends a follow-up question, the entire flow is triggered again.
@@leonvanzyl Do you provide any type of consultancy or mentoring in the development of these solutions? I'm very interested.
Can you share with us the flow agents diagram in json please ❤❤??
Ok, will see what I can do
Thanks!!
How can the agent ask for human feedback in the loop, such as approving the essay?
Hello, I'm new to this, but after watching this great tutorial twice, I would suggest inserting a "condition" node right after the "start" node, similar to the "hotel" example with 3 agents. The condition to evaluate would likely be something like "detect if the user is disapproving the essay produced," with results "yes" or "not," and then ending the conversation or redirecting the flow to the "3 loop essay production" again.
However, I think it might be necessary to consider a state variable first to determine if the essay loop has already been run at least once. Or something like that 😅
It certainly is a new way to develop conversational flows... super We're all new here.
Furthermore, I think I have also seen that within a "chatflow," we can call other "chatflows." I suppose that with agents, it is probably possible to define some kind of "agent orchestration." Then there should be the "3 loop producer," hehehe, which would be called by another agent or by the "orchestrator" agent (I think this is a "supervisor" agent).
Man, we all need to dedicate a few hours to play with this!! 😅
Sad that it does not work, after adding the llm node after the condition node, i don recieve an text from the node behind anymore. So my Ai does not ask me for my name anymore.
You might be missing a few steps from the sound of it. The LLM node by itself won't produce much in terms of a response. Did you follow the video all the way to the end?
@@leonvanzyl yes i did, 3 times alredy, step by step, before i add the llm, my agent ask me for my name after introduce the llm. It dont ask anymore, i just get the empty Response
@@leonvanzyl I've now found the problem. I tried to replicate your tutorial, with llama3.2 via ollama. I tried nearly everything, but the last thing I haven't tested was the possibility, that a locally hosted model, would be not able to do the simple tasks asked from it. So I paid a few bucks for the openaiAPI and tested with that... It works!! I'm a bit sad tough, I would have loved tho get that thing working fully locally...