Core concepts
The need-to-knows of EB life
Building AI Agents with Easybeam is a breeze, but it can still be overwhelming to translate the ideas you have in your head into an actual AI Agent.
Let's go through a couple of core concepts that'll help you on your AI Agent journey.
The AI Agent flow you build will process each message.
Every time you send a new message to your AI Agent, the entire flow will be triggered from start until it ends in a Prompt Step. This means you're essentially creating a whole brain that's enhanced with actions, rather than a flow-chart of how to respond.
Easybeam is stateless (unless otherwise specified).
Your agents work by processing the entire list of messages sent to them to generate a response. There is no memory required to do this, so although an interaction will seem like an entity is learning and developing a memory, it's effectively an illusion since nothing is stored.
The exception to this is if you want to build an AI Agent that does have a long term memory. To do this you can leverage the Data Step to create a database of previous interactions that can be queried to create a "long-term memory". Check out the instructional video here.
Each flow must end in a Prompt Step.
This is so that you can stream in the final response from your AI agent. Our also allows for non-streaming responses, but for maximum user-joy we recommend streaming the completion in as it occurs.
If you want things to happen after your final step, we recommend using an Async Step just before the final step to trigger any cleanup/update work you might require.
Last updated