LangChain Agents : A VERY SIMPLE & Quick Tutorial!

Key Takeaways

[Music] hello in this video we talk about agents in Lang chain suppose that you want to ask a question from a chatbot like chat GPT to know the temperature of a city like New York however our chatbot may not access to real-time data or it may have been trained on old data so its knowledge may not be up toate using langing chain agents we can take an action such as a web search to provide relevant data for our chatbot to provide more useful answers here's a simple flowchart of Lang chain the user may directly ask a question from a specific large language model such as GPT Lang chain supports various language models the user can also connect its own data to the language model through Lang chain to get more relevant or customized answers from it for example he can connect his own PDF files to the language model as an external source of information after getting the response of the language model we can use Lang chain agents to take an action or make a a decision based on the generated response for example the agent can call an external API to get more information to be used by the language model this can be done by Lang chain tools Lang chain supports several tools such as Google search python shells terminal commands Wikipedia search Sur API and many more we next show you how to design a simple agent to do a search task using Sur API Sur API is an API that scraped Google and other search engines such as Bing now let's write a simple python code in Google collab to make a Ser API agent first you need to install the Lang chain package this can be done by the PIP command we then need to import three modules from Lang chain so that we can load the tools initialize agents and also Define an llm such as open AI using this command you can load the GPT model from open AI you need to put your API key here if you set the temperature parameter to zero you will get more deterministic or less random answers We Now define a list of our tools and load them here we just use the Ser API as our only tool after that we create an agent and initialize it using the loaded tools and our chosen llm the zero shot react description is a generic type of the agents and you can use it in many tasks finally we run the agent by feeding our question or query as an input to it the agent get the response from Ser API and feeds it to the llm to get our answer we can also use a specific wrapper for many tool and L chain for instance we can use a wrapper for Ser API we then create an instance of it and feed our question or query to it to get the final response moreover we can specify some custom parameters for each rapper for example we can specify the search engine the location the language or other parameters Lang chain supports many tools the list of the supported tools are growing rapidly you can check the official website of Lang chain to access the complete list of such tools with their documentations you can also build your own custom agents and tools we will show it in another video