Docs / Tetonic Engine
Architecture
The engine separates how work is requested, coordinated, executed, remembered, and connected to models.
Applications and interfaces
Applications direct the runtime through interfaces. The source tree includes terminal and daemon interfaces. Existing executable names such as lokai and lokaid may still appear in build instructions; they are interfaces into the engine, not separate product lines.
Coordination and execution
Coordination manages work across agents and execution resources. The runtime carries out agent work within configured tool and policy boundaries. Keeping these responsibilities separate is central to the distributed design.
Durable state
The state layer holds execution history, memory, and artifacts. This gives longer-running work a record beyond the current model context. Retention and recovery behavior should be evaluated for the implementation and deployment you use.
Model access
The inference layer connects the runtime to models. Model selection, network access, and available compute are deployment decisions.
Find it in the source
The repository uses five layer names: Litho for interfaces, Mantle for coordination, Core for execution, Strata for state, and Atmos for model connectivity. These are internal architecture names within the Tetonic Engine.