# Coverage demoted 2026-10-07 (roadmap#942, PLAN.md Stage A): this file was written by an API Evangelist # sweep and names no billing unit, meters or pricing category read from the provider — it restates the # FinOps framework. It is `x-coverage: unmeasured`, and the FOCUS alignment it claimed has been removed. x-coverage: unmeasured specification: FinOps Framework specificationVersion: '1.0' provider: Jentic providerId: jentic publisherName: Jentic serviceCategory: AI Agent Platform created: '2026-05-04' # Provenance stamped 2026-08-11: this artifact was written by the API Evangelist # bulk sweep dated 2026-05-04, not harvested from the provider. See roadmap#35. method: generated modified: '2026-05-15' reconciled: false notes: >- Jentic's Enterprise Edition pricing is gated (contact sales), and the Core Edition is free, so a published unit-rate card is not available. Meters and unit economics below are aligned to the obvious chargeable signals on the platform (workflow executions, integrations used, agent runs, seats, MCP calls). Reconcile against a real Jentic order form before treating rates as authoritative. tags: - AI Agents - Arazzo - OpenAPI - MCP - Workflows - FinOps - Cost Management - FOCUS description: >- FinOps framework definition for the Jentic agentic knowledge layer. Provides a FOCUS-aligned mapping for cost allocation, usage measurement, and unit economics across the search / load / execute surface, the Remote MCP server, and the Agentic Sandbox. principles: - name: Visibility description: >- Make AI agent consumption costs (per agent, per workflow, per upstream API) visible to engineering, product, and finance in near real-time via execution_id telemetry. - name: Allocation description: >- Every chargeable agent call is tagged with the agent UUID, scope, upstream api_name, and execution_type (operation vs workflow) so cost can be split across teams and features. - name: Optimization description: >- Continuously evaluate which goals should be Arazzo workflows (cheaper, deterministic) vs free-form tool calls (more LLM tokens), and which APIs are pulling disproportionate cost. - name: Accountability description: >- Establish per-agent budget owners and chargeback or showback flows. Enterprise scopes can be aligned to business cost centers. domains: - name: Understand Usage and Cost capabilities: - Data Ingestion - Allocation - Reporting and Analytics - Anomaly Management - name: Quantify Business Value capabilities: - Planning and Estimating - Forecasting - Budgeting - Benchmarking - Unit Economics - name: Optimize Usage and Cost capabilities: - Architecting for Cloud - Rate Optimization - Workload Optimization - Cloud Sustainability - Licensing and SaaS - name: Manage the FinOps Practice capabilities: - FinOps Practice Operations - FinOps Education and Enablement - Invoicing and Chargeback - Onboarding Workloads - Intersecting Disciplines billingModel: pricingCategory: Subscription Plus Usage billingFrequency: Annual billingCurrency: USD chargeCategories: - Subscription - Usage - Purchase - Tax - Credit - Adjustment chargeFrequency: Recurring focusColumns: ServiceName: Jentic ServiceCategory: AI Agent Platform ProviderName: Jentic PublisherName: Jentic InvoiceIssuerName: Jentic PricingCategory: Subscription Plus Usage PricingUnit: execution BillingCurrency: USD ChargeCategory: Usage meters: - name: workflow_executions description: >- Count of /agents/execute calls where execution_type = workflow (Arazzo workflows). unit: execution aggregation: sum dimensions: - agent - workflow_uuid - api_name - status - tier - name: operation_executions description: >- Count of /agents/execute calls where execution_type = operation (single OpenAPI operation). unit: execution aggregation: sum dimensions: - agent - operation_uuid - api_name - status - tier - name: integrations_used description: >- Distinct upstream APIs (api_name) actually invoked by an agent during the billing period. Counts breadth, not depth. unit: api aggregation: count_distinct dimensions: - agent - org - tier - name: agent_runs description: >- One end-to-end search → load → execute session, correlated by execution_id chain. unit: run aggregation: sum dimensions: - agent - tier - name: seats description: Active human operators on the Jentic console. unit: user aggregation: max dimensions: - org - role - tier - name: mcp_calls description: >- Tool invocations against the Remote MCP server at api.jentic.com/mcp (search_apis, load_execution_info, execute). unit: call aggregation: sum dimensions: - agent - tool - client - tier - name: search_queries description: Count of /agents/search calls. unit: query aggregation: sum dimensions: - agent - tier - name: load_calls description: Count of /agents/load calls. unit: call aggregation: sum dimensions: - agent - tier - name: data_egress description: Bytes returned over the network in /agents/execute responses. unit: GB aggregation: sum dimensions: - agent - api_name apis: - name: Jentic API baseURL: https://api.jentic.com/api/v1 serviceName: Jentic API serviceCategory: AI Agent Platform tags: - AI Agents - Arazzo - OpenAPI - Workflows - name: Jentic Remote MCP baseURL: https://api.jentic.com/mcp serviceName: Jentic Remote MCP serviceCategory: MCP Server tags: - MCP - AI Agents unitEconomics: - name: Cost Per Workflow Execution metric: billed_cost / workflow_executions - name: Cost Per Agent Run metric: billed_cost / agent_runs - name: Cost Per Integration Used metric: billed_cost / integrations_used - name: Cost Per Seat metric: billed_cost / seats - name: Cost Per MCP Call metric: billed_cost / mcp_calls maintainers: - FN: Kin Lane email: kin@apievangelist.com