Installing tools via packages

What are packages?

Packages are collections of tools that you can use to extend your agent like generating images via OpenAI, retrieving Stripe customers or getting the current weather.

Technical details

Each tool is a function that takes arguments a.k.a. request parameters and returns a response. You can read more about packages on Package overview, but in the most basic sense packages are a group of related functions will take some input and return an output.

fn(input) = output
tool(argA, argB, argC) = response

e.g.
get_weather(lat, lng) = { temperature, wind_speed }
get_customers(count) = [ { name, age, payment_method }, ... ]

Packages are standalone REST HTTP API servers. They run on auto-scaling serverless architecture via AWS Lambda behind an Instant API gateway that performs type-validation and coercion automatically.

How do I install packages?

You can manage your agent's installed packages via the Packages tab in the top toolbar on your agent's chat page. When you click this tab you'll be able to see your installed packages.

By default your agent won't have any packages installed. In this case you can select the Discover sub-tab to find suitable packages.

The Discover tab will only show approved packages, that is, packages that have been vetted by the Instant.bot team for security and authenticity purposes. Before we install a package, we can check the package details by clicking on the name or by clicking the eye button to view individual endpoint settings.

For the current weather package, we can click the eye button to see which endpoints it has before installing. From here we can toggle individual tools on or off.

For this package, there is only one tool. We want it enabled! So let's go ahead and click [ + Install ]. You should see a popup like below:

That's it! You should now be able to use the current weather package.

Using installed packages

Once a package is installed, your agent will automatically detect when to use it based on your requests. For example, for the weather package above, we can ask the agent something like:

What's the current weather in San Francisco?

And your agent should go through two steps:

1. Planning: your agent determines what tools it has available and if it should use any to answer your query.

2. Execution: your agent calls any and all functions necessary to complete your request.

The result should look something like this:

You can click on either the Thought [...] button (in this case, Thought for 2.35 seconds) or the Function [...] button (Retrieves current weather...) to expand details and see how your agent thought about this task and executed on it:

You can use these buttons to debug your agent and figure out which packages and tools work best.