Two Ways to Save 96% of Your Money Using DeepSeek R1 in n8n

[@alexflint1941]

I am over 55 but because of you, I have been able to build an AI agent (and understand it)! Thanks so much for your help Nate :)

[@ffmdotcom]

Not sure deepseek-r1 is a good fit for n8n (At least for regular agent stuff). I mean it’s a fantastic model but it has pretty "firm principles" which could make it difficult to use for a wide range of tasks. And it certainly doesn’t make for a great "chat buddy" either. But I love it's open process which comes with its very own charm. Generating code and solving puzzles sure works pretty good as well.

[@aaronabuusama]

It’s not supposed to. You should think about reasoning models as tools for background tasks while “regular” models are for interfacing with the user or tasks where the reasoning is determined by the programmer Given the fact that Ilm’s are better at prompting llm’s, the way I see these being used in agent systems is having a regular model as the orchestrater kicking off async flows that may take a while to complete

[@ffmdotcom]

@@aaronabuusama I agree with you - I just think that the way DeepSeek R is pictured is a little too general. It's great but you can't just plug it in an replace any other model. One has to understand what it really is. It might indeed be a good orchestrator given it's reasoning it should be good at breaking down whatever is thrown at it.

[@aaronabuusama]

@ but it’s not a general model. It’s a thinking model. It’s not supposed to be a drop in replacement for any other llm

[@ffmdotcom]

@@aaronabuusama correct - it's not a general model...hence I made my initial comment.

[@brianhinton978]

Can you please tell me how you just solved my 72-hour issue in like 45 seconds? 7:02

[@sfl1986]

The issues with function calling are not coming from open router. It's from deepseek. They are saying it's highly unstable for now on their own documentation for function calling. One setup I'm thinking of trying to get best of both worlds on a chat agent is having a deepseek R1 basic llm chain at the begining of the initiated chat flow to come up with the best plan as fast responding to the query weather or not to use tools and which tools and how. Then after crafting a well detailed plan for the response pass it on to a full ai agent node powered by 4o mini which is reliable for function calling (or Gemini 1.5 flash) and have a very simple prompt that makes it act (not think) on the R1 plan to execute functions when needed then send back the response to the user

[@webscrapingwithola]

Thanks for the video, the link to join the community is not working, it only leads to the discovery page, please update the link, thanks

[@nateherk]

You will join the community from that link!

[@bangkitsanjaya3773]

Is Deepseek didn't have base url?

[@nateherk]

Check out my latest video to see how to access it through DeepSeek’s base URL: kzbin.info/www/bejne/qpvEdWlrhKiie5o

[@kanchanapathirage]

how to setup for tool calling?

[@nateherk]

It's not really optimized for that yet... I found a cool way to use it for making your agent tool calling more efficient though... kzbin.info/www/bejne/qpvEdWlrhKiie5o

[@giladbar-ilan1472]

thanks!! i am doing the same based on the json you shared, though i am getting 401 - "{\"error\":{\"message\":\"Authentication Fails (no such user)\",\"type\":\"authentication_error\",\"param\":null,\"code\":\"invalid_request_error\"}}". any idea why?

The model selection doesn't work if using Deepseek's API base url, unfortunatelly. :(

[@nateherk]

Seems like OpenRouter has been having some issues!

@nateherk oficial Deepseek API too then, I was talking about that.

[@Vaultix123]

Great job! You are very detailed. Could you please explain some alternative routes to integrate WhatsApp without the need to have a verified business account?

[@mchu7371]

why do you like n8n instead of creating your own scripts using ai?

[@coded4348]

I hope deepseek will not become deepshit in the future.

[@GiZmo-zz]

Could you do video about sending lets say like 500 emails with gmail ONE BY ONE from QUERY with loop system, so like it loops the query of emails and sends them one by one looping it. without sending them to same persons that it already sent.