Skip to main content
Building with StateBase requires a shift in how you view AI. Most developers view an agent as a Linear Function: Input -> LLM -> Output. StateBase developers view an agent as a State Machine: Current State + Input -> LLM -> Output + New State.

​
1. The Session as a Ledger

Stop thinking of chat history as a list of strings. Think of it as a ledger of transitions. Every interaction is a change in the agent’s internal world. If you have the ledger, you can reproduce the world.

​
2. The Context is a Construction

Don’t pass “Everything” to the LLM. Construct a “Prime Directive” for every turn.
  • Who is the user? (Memory)
  • Where are we? (State)
  • What just happened? (Turns)

​
3. The Model is a Controller

In the StateBase world, the LLM is the “CPU,” but StateBase is the “Bus.” The LLM decides what to do, but StateBase ensures it has the instructions and remembers the result.