0:03 AI agents only work when they have 0:04 access to the right set of tools. 0:06 Luckily, when you're building a retool, 0:08 there are a few different ways to give 0:09 your agent access to the tools it needs 0:11 to be productive. Let's take a look at 0:13 how that works. The tools section lives 0:15 at the bottom of the configuration tab 0:16 for your agent. Clicking on the add tool 0:19 button shows you all the options you 0:20 have available. As you can see, retool 0:23 gives you access to a library of 0:24 pre-built tools that you can add to your 0:26 agent with a single click. If you have 0:28 existing workflows built in retool or 0:29 tools built for another agent, your 0:31 current agent can use those tools as 0:33 well. You can even use an entirely 0:35 different agent as a tool. This allows 0:37 you to build agents to do specific work, 0:38 but still enables you to create complex 0:40 multi- aent systems that can work as a 0:42 team. And as more and more third party 0:45 services spin up MCP servers, you can 0:47 link those servers to Retool to let your 0:49 agent do real work in your other 0:50 business systems. 0:52 Finally, if you need a tool that's not 0:53 covered by any of these options, you can 0:55 create your own custom 0:56 tool. When building on a custom tool, we 0:59 first need to give it a name and a 1:00 description. We can also decide if we 1:02 want to ask the user for confirmation 1:04 when the AI has chosen to use this tool, 1:05 but before it runs. This is useful as a 1:08 double check on tools that perform 1:09 irreversible operations like refunding a 1:12 customer or writing to a database table. 1:14 Then we define what parameters the tool 1:16 expects. This is how our agent knows 1:18 what information is necessary to call 1:19 the tool. These parameters can either be 1:22 required or 1:23 optional. In this case, we'll build a 1:25 tool that constructs a greeting for a 1:27 user. So, we can ask for their name as a 1:28 parameter. You can adjust the maximum 1:30 time your tool can run before timing 1:32 out, which is useful if your tool is 1:34 calling a longunning API or takes longer 1:36 than 10 seconds to execute. And finally, 1:39 we can build out the function that 1:40 powers our custom tool. Clicking the 1:42 edit function button takes us into the 1:44 function editor where we can lay out all 1:46 the blocks our function needs to run 1:47 properly. In this case, we'll add a code 1:50 block that takes the name parameter and 1:51 adds a greeting before returning 1:54 it. Now, when we ask for a greeting in 1:56 our agent's chat, we see that our tool 1:58 is called and the user is 2:00 graded. And that's it. Hopefully, your 2:02 head is spinning with ideas of tools to 2:04 connect to your agent, whether they 2:05 already exist or you need to build them 2:07 yourself. In a future video, we'll take 2:09 a look at how to make sure your agent is 2:10 calling the right tools and working as 2:12 expected. See you next time. 2:16 [Music] 2:17 [Applause]
