The philosophy of Kōhai

Kōhai is the AI assistant embedded in Jinkō. It helps users work with the platform's evidence, models, Virtual Populations, Protocols, Trials, and results. The underlying Jinkō workflows remain available directly through the interface, SDK, and REST API.

See the Kōhai overview for the assistant's current capabilities.

A junior partner

Kōhai (後輩) means “junior” in Japanese. A kōhai learns with guidance from a senpai (先輩, “senior”). The name reflects the intended working relationship: Kōhai accelerates defined tasks while the expert sets direction, reviews the work, and accepts the scientific result.

Users choose how much of a workflow to delegate to Kōhai.

Bounded actions

Kōhai works through operations available in Jinkō. This keeps agent-assisted work within the same access controls and project structure as work performed directly in the interface.

• Suggestions correspond to existing Jinkō operations.
• Consequential actions require the level of confirmation configured for the workflow.
• API errors and validation failures remain visible to the user.
• The user's permissions bound what Kōhai can access and change.

Traceability and review

Agent-assisted work should be inspectable in the project where it will be used. Jinkō records new item versions so users can compare changes, review inputs, and return to an earlier version when required.

Version history identifies agent-assisted changes and supports review.

The project team remains responsible for validating model assumptions, parameter values, units, data transformations, simulation design, and interpretation.

Data handling

Kōhai uses project data to perform the requested workflow. User-specific data is not used to train Kōhai. Access follows the platform's project permissions and security controls.

For the broader standard applied to internal and external agents, see Ownership and review.