Skip to main content

Jason-like Kotlin Agents: where JaKtA comes from

· 2 min read
Samuele Burattini
Samuele Burattini
Maintainer

JaKtA started as a new implementation of Jason, the interpreter for an extended version of AgentSpeak(L), written in Kotlin. The name was originally an acronym for Jason-like Kotlin Agents. With 1.x it no longer is.

JaKtA 0.x: Jason-like agents in Kotlin​

JaKtA 0.x brought AgentSpeak-style agents into Kotlin, as an internal DSL instead of a separate language. It followed Jason closely: beliefs and goals were always Prolog terms, goals were either achievement or test goals, and agents acted on an environment through actions. The aim was to explore paradigm blending: writing BDI agents with the same syntax and tools as ordinary object-oriented and functional Kotlin code.

JaKtA 1.x: moving away from Jason​

JaKtA 1.x keeps the BDI model, but is no longer an AgentSpeak(L) implementation:

  • Knowledge representation is generic. Beliefs and goals are not tied to Prolog terms: an incarnation chooses them, and also defines how plan triggers and guards match.
  • Plan bodies are Kotlin. They are ordinary suspend functions instead of AgentSpeak body formulas, and test goals are replaced by guards.
  • The engine is coroutine-based. Each intention is a coroutine, see the execution model.
  • Agents live in nodes, not in a Jason-style environment, and act through skills instead of environment actions.

Jason is still an influence​

The Prolog incarnation supports belief sources and KQML messaging in the style of Jason, which makes porting Jason programs straightforward (see the Blocks World showcase). But it is one incarnation among others, and today JaKtA is just a name rather than an acronym.