1. How do agentic automations work?

CODEXIS AI supports automations that run an agent either manually or on a schedule. Each automation primarily contains a name, a description, a prompt, a selected agent, a set of skills, a working directory, a step-count limit, and a schedule in the standard five-field Linux cron format. When triggered, the system invokes the selected agent, passes it the assigned skills, and performs the run over the specified working directory.

Each run has its own history and status, for example RUNNING, COMPLETED, or FAILED, so you can trace back what the automation did and with what result. Automations are stored separately for each user within the runtime environment. If a new time-based trigger fires while the previous run is still in progress, the system does not build an unlimited queue but holds at most one pending run. This limits the uncontrolled accumulation of tasks.

In practical terms, this means you can set up, for example, nightly document processing, regular reports, review workflows, or recurring integration tasks without the need for manual triggering.