← Back to journal
Architecture / July 19, 2026

How Termyte Processes Agent Work

Termyte separates captured activity from observations, memories, and the context later sessions receive.

First, record the activity

Connected integrations send agent events to Termyte. Termyte redacts configured content before persistence and stores the resulting trace records in its local SQLite database.

The capture path is designed to finish quickly. Background jobs handle the heavier processing afterward.

Then, preserve the links

Background processing can turn related traces into observations and then into reusable memories. Those derived records retain links to their source observations and traces, along with repository, session, file, and task context where available.

This separation matters. A memory is not presented as an unexplained answer. The viewer and explainability tools can show where it came from and how it was selected.

Finally, build context

Termyte ranks eligible context within a token budget. It records the context packet and the items shown to the agent. Later feedback can be attached to that delivery.

The result is an inspectable continuity path:

capture → evidence → observations → memories → context → feedback

Each stage has limits, but each stage exists in the current product.