Claude Managed Agents: керовані агенти від Anthropic

Clude managed agents a direct competitor to open claw but a managed version. Here is the managed agent in cloud console having the ability to send message to my slack and running 24/7. Integrate with any of your application with detailed traces work is changing. Previously AI answered questions but now AI does the work. Your agent has an identity, a name, a role, and a home. That is the environment. Previously, it generates text about the work. Now, it does the work, writes and run the code, execute the steps, delivers the output, choose what tools to use, block what is sensitive, and stop in the mid-flight. One agent in finite sessions. You can continue from where you left off. It remembers. Today I'm going to take you through step by step how you can create cloud managed agents through its own user interface and how you can programmatically integrate this with your own application also with just few lines of code and going through its different features. All you need to know about managed agents that's exactly what we're going to see today. Let's get started. >> [music] >> Managed agents are nothing but a agent having its own environment with tools and then session to remember the previous conversation. So these three is available in open claw. But how this differs from open claw manage agents is anthropic hosted. Open claw you can self-host. The key difference is the infrastructure. To host open claw you need to know technical information. You need to understand how to host the application and you need to find a server to run it and that is tedious. But using this managed version all the infrastructure work is been handled by anthrop. Ultimately as a user you want your agent to perform task. So you spend time making your agent to perform task rather than focusing on building the infrastructure. So managed agents removes the headache of managing your own infrastructure that is setup and configuration. But with open claw you need to manually do it. So now we're going to see how you can set a managed or create a managed agent. There's a clear documentation on how to create managed agent has a prerequisite. You need anthropic install using pip install and then exporting your anthropic API key. Anthropic API key generate from platform.cloud.com. So I'm in cloud console there. You can generate your API key here or you can click this icon to get your API key. So generate your API key and save it somewhere secure. First I'm going to show you how you can integrate this with your own application. Next I will show you how to create this manage agents through the user interface. So first I'm going to install it from your computer. Open terminal and then click enter. So there pip install anthropic and then click enter. That is the main package. Once that is installed, export your anthropic API key. This is the one which you have just generated. Put that here and then click enter. Next, I'm going to create a file called app. py. So, I'm creating it like this. If you use VS Code or any code editor, you can even right click and create new file app. py and then click enter. So, here we're going to write the code and using the terminal, we are going to run it. So, first from anthropic import anthropic then client equals anthropic. First we are going to create an agent. Agent equalsclient beta agents create and then I'm creating agent with the model name system prompt and the tool set. So by using this agent tool set we are enabling the full tool set which includes all these tools and that's it. That is how you create an agent. Next we going to update the agent. Once after creating the agent if you want to update any of these information you can use this function dot update and that will update the name and the system prompt. So the three key important steps involves creating an agent and creating environment and third creating a session. So next we're going to focus on creating an environment. So step number three creating an environment. So I'm going to use environments.create create function and defining the environment name the config going to using the cloud and unrestrict and that's it the environment got created and the third important step is creating a session so next create a session using sessions create function and here I'm providing the agent ID and the environment ID I got the environment ID from the previous step and also agent ID from the previous step now finally we are ready to talk to the agent number five stream a response So to do that I'm going to use sessions.events.stream function. So I'll provide all the code in the description below for you to try it out. So here to the agent I'm going to ask a task to do write a Python script that prints hello from managed agents and run it. So that is a task I'm going to give to the agent. So that is the user message and the task is provided in the content key. So once that event is sent, we'll be printing the response. So this is how you send any task to any agent by using this code and then you print out the response by using this a for loop and then it prints out the result. So now we're going to see how it's going to perform. I'm going to run this code. So I'm going to go back to my terminal to clear it and python app. py and then click enter. And now it's running. You can see agent updated environment created and here's the session ID created. Now it's streaming the response. So it's trying to write the Python code what we asked for and it used the bash script and it saved the result here in this path. So that is the path within the environment that is anthropics managed environment. The same process these much amount of code up to line number 73 can be compressed into just few lines using praise and AI and you got only four lines and I'll show you towards the end how to do this. But let's continue with our application. You can ask question back to the agent which you just created. You can track the usage. You can list the sessions. You can select tools, disable tools, create custom tools, perform a web search, use packages and also you can stop in the middle of a conversation. So here is the multi-turn conversation. So same like before we are sending session events and now we are asking modify the script which was initially generated and then printing same as before. Next, to track usage, you can use usage.input tokens and usage.output tokens from the results. To list the sessions, you can type list sessions.list. If you want to use only few tools, not every tools, then you can define that in the tools configuration when you create the agent. So that is the first step we did. We created the agent. During the process, we are defining what are the tools to enable such as bash, read and write. Next, if you want to disable some tools, you can just mention enable equals false and that will disable these tools. If you want to create custom tools, you are defining the schema of the tool. I've already covered about creating a custom tool for CL which I'll put the link in the description. So, this is the schema and the tool name is get weather. So, now when you ask the agent what is the weather in Tokyo, it'll automatically try to use the custom tool with the parameters. In this case, it's Tokyo and it'll return the result. By default, to enable web search, it included this in this tool set. You can also make this web search to true to enable it. If you want to include pre-installed packages such as pandas, numpy, you can include that in your environment. So, when creating an environment, you define the list of packages required. And finally to interrupt a session if a message has been sent you are stopping it by like doing like this end is equal to empty. And if you want to stop a tool call then you are using this type equals user.in interrupt. If the session is already ideal then you just skip. So that is a quick summary of these all features which I'll put the code in the description below. So now I'm going to run this code the whole code which automatically creates the agent set up the environment create the session update the agent create custom tools and then interrupt the session. So coming back to my terminal python app py and then click enter and now you can see agent is getting created agent updated environment created session created and it's streaming the response the first step of creating the python script. Now the second step of updating the existing script which it created before that is multi-turn so you can keep on conversation with this agent. Next we got the usage report input tokens output tokens then total list of sessions. I've got many sessions right now. Next allowing only the bash tool and you can see the response here. Next disabling web search. Next using a custom tool to get the response and you can see here. Next, enabling the web search and searching for relevant information. Next, we are installing pandas and numpy into that environment. So, a new environment we are setting it up and now it ran the code with pandas as you can see here data frame. And finally, interrupt and it's interrupting in the middle. So, that is really nice. Next, I want to show you to do this in a more simplest way. So, for this purpose, I'm going to do pip install praise AI and then click enter. that will automatically install the praise and aai package. So only these few lines of code from praise import agent and managed agent and here I'm using a back end that is managed agent without this back end it'll automatically run this code locally but with this back end it's going to use anthropics managed agent so now I'm going to run this code but going to give the same task as before to create the Python script and print the results and going to my terminal back again Python and that code and now it's running and here is the result I've created and executed the Python script here's what I did And here are the results. So with just few lines of code you are able to do the managed agent. Similarly to update an agent manage.update agent the environment and session is automatically created behind the scenes. So you don't need to do anything about that. To stream the response just at the end add stream equals true for the agent. method. Manage. session will retrieve the session. Manage list sessions will retrieve the list of sessions. You can customize the list of tools to add and remove and the custom tool that is weather can be easily used here. You just define the schema and nothing else and everything works as it is. Next, I want to show you how you can use this in the user interface that is anthropic own console. Go to platform.clude.com. There you got managed agents. So as a quick start, you got various steps here. The same thing what we saw before. create agent, configure environment, start the session and then integrate. So there are some templates here. These are simple YAML files. If you want deep research, you can click that and it is a YAML file with all the required steps to take with the required tools. But for my case, I'm going to use notion and Slack integration support agent. Clicking on that. So that is the template settings. Here I'm going to click use template. And on the left hand side you can see automatically it says agent created. So the settings what we saw here that is the settings for the agent. By clicking that you create the agent. Next we need to configure the environment. So click configure environment. Now it's asking me a few question. For what purpose or do you want to reuse the existing environment? In my case I'm going to say create a new environment. What level of network access do you want to allow this agent to have web access or not or to restrict any external tools? In my case, I'm going to say unrestrict. Now it's creating the new environments. Environment created. Next, start the session. Clicking on start session. And now the session is running. Next, I can click Slack and it'll automatically take you to Slack page and click connect. I've already done that. and similarly to notion. So automatically all your credentials get stored in credentials vaults on the left hand side of the console which you can view later. So I'm going to choose the vault I want to use because I've already configured the vault before of connecting with Slack and notion and clicking on support agent vault. I own or I'm authorized to use this vault. Confirm. So in my Slack I've got a channel called support. So I'm going to click that and support internal for internal communication. Now I can run my test test run. So on the right hand side you can see I got my agent ready with the session and environment created. So to test this I'm going to say how do I reset my password and clicking send. You can see on the left hand side what the agent is trying to do for clarity purpose. And here you can see the step by step tracing. So all the tracing is included in this application. So every single step it's taking. Agent is thinking, searching the knowledge base, then it found the chat history and then sending a message to Slack with the channel ID. And let me open my Slack. And here is the response. Great question. And at the bottom, you can see send using clude. As simple as that. Now using programming or the coding, you can interact with this agent which we have just created through the user interface. I'm going to click the view session. So you got the session ID. Using this, you can call this agent. Again you can see the process or progress here. Once after you're happy with your work you can stop the session by clicking this icon and the session is now closed. If the session is open that means you pay for that session and the final step is to integrate and it automatically gives you the code the agent ID the environment ID to interact with the agent which we have just created. So this is using the user interface. So ultimately the final step it's taking towards is integration using the Python code. using the user interface. If you're creating one agent, that is fine. If you're creating two or three, that is fine. But what if you want to create hundreds of agents, that's when you need the initial programming steps which I explained. So these steps are exactly same as what we did in the user interface, but much more defined and it's easy for you to integrate this with any of your application. Do try and let me know in the comments below what you think about this. Considering you already like Claude managed agents, I also created another video about Claude custom tools, how to create that. I'll put the link in here and I highly recommend for you to watch and I will see you