Skip to main content

Terminal

Terminal runs a named job against a pinned subject, and returns a research document with its sources attached. Both halves matter. The job is chosen from a closed catalogue rather than described in prose, and the subject is a canonical identity resolved before the run starts, not a company name the model has to guess at later.

The job catalogue

company.financials and comparative.research accept structured configuration (which metrics, over what period, and which table columns) rather than encoding that intent in free text. See Definitions.

Pinned subjects

A subject is pinned by canonical id, with the evidence of how it was matched carried alongside it. A workflow that runs unattended requires an exact match, a ticker or an exact name, because a fuzzy match that drifts to the wrong company is worse when no one is watching.

Outcomes are not binary

A job can return complete or partial research. Partial means at least one part of what was asked could not be evidenced, and it is common in some job classes:
Measured across production runs on 2026-08-09. market.lookup and company.research are systematically partial today. A workflow built on them will usually produce partial research, which the default delivery policy holds rather than sends.
This is why a run reporting completed tells you the machine finished, not that the research is whole. The two are reported separately. See Runs.

Not yet in the API

Terminal is not separately addressable over the API. It is reachable through the web application, and as the steps a workflow is composed of.