Glossary
Use these definitions when following the task guides. Your access and organization’s configuration determine which capabilities are available.
| Term | Meaning |
|---|---|
| Workload | An AI capability such as an assistant or classifier, with its own owner and configuration. |
| Asset | A reusable building block, such as a prompt, flow, model configuration, or policy. |
| Asset version | A specific immutable record of an asset’s content. Releases select particular versions. |
| Flow | The connected steps that describe how a workload performs its work. |
| Release | A candidate or activated configuration tying a workload to specific asset versions and settings. |
| Binding | The association between a release and an asset version or other required configuration. |
| Policy | Rules that govern whether and how an action may proceed. |
| Evaluation | A test or assessment of behavior within a defined scope. Passing an evaluation does not prove every possible behavior is correct. |
| Evidence | Recorded information used to investigate execution or decisions. Its coverage and visibility depend on the operation and your access. |
| Readiness | Whether the required conditions for a workload or release are satisfied. Registration alone does not establish readiness. |
| Sandbox | The configured operating limits and permissions used to constrain execution. |
| Service lane | The execution service assignment used by a workload. It is separate from a release’s promotion or rollout lane. |
| Workspace | The team context in which you access workloads and other resources. |
| Namespace | An ownership and cost boundary grouping workspaces within an organization. |
| Organization | The customer or organizational boundary in an Orqestra deployment. |
| Connection | Saved configuration connecting a workload to an external service in a workspace and environment. |
| Secret | A protected credential used by a connection or service. |
| Memory | Stored information the assistant may use as context in later interactions. |
| Context provenance | A record of context considered for a response, subject to access and redaction. It is not private model reasoning. |
| Cohort | One selected configuration or candidate included in a comparison. Compare evidence applies to the chosen cohorts and dataset. |
| Dataset / evaluation suite | The selected test cases and scoring rules used for an evaluation or comparison. |
| AI Estate | The available inventory and impact views for AI systems, with explicit coverage and access limits. |
| Concurrency | The number of executions allowed to run at the same time within a capacity pool. |
| TPM | Tokens per minute: a provider/model rate limit, separate from concurrent executions and spending limits. |
| Circuit-break | A temporary control that blocks affected operations; inspect its scope, reason and expiry. |
| Human decision | An authorized person’s approval or denial of a specific pending action, separate from that action’s execution outcome. |