| AI FinOps Capabilities |
| Focus | Governance: access, guardrails, budgets, tool policies, and spend attribution on every request | Observability, with an AI gateway and request logging |
| Policy on every request | No policy, no traffic: a request without a policy you bound is refused with a clear reason | Not a policy-first gateway |
| Guardrails | Built-in credential detection and personal-data rules, plus Azure AI Content Safety, AWS Bedrock Guardrails, Google Model Armor, a webhook, or Cloptima-managed scans | Prompt-injection and jailbreak screening, with optional content moderation |
| Personal-data protection | Personal data and secrets are redacted before the provider sees them, then restored to the original text in the response, including streamed responses, tool calls, and structured output | Not its focus |
| Budgets | Daily and monthly budgets per team, app, environment, or key, one org-wide ceiling, Alert only or Block, held before the provider call | Rate limits |
| Org baseline policy | One organization-wide baseline for models, tools, guardrails, limits, and budgets. Every request gets its own policy plus the baseline, and the stricter setting wins, so teams can tighten the rules but never loosen them | Not its focus |
| MCP and tool governance | Allow and deny lists for tool servers and tools (denied tools are taken out of the request), approvals before a server goes live, tool definition changes flagged, and spend by tool, set per policy on top of your org baseline | Not its focus |
| Response caching | Policy-scoped exact and semantic cache, both with Monitor only before Serve | Exact-match caching |
| Sessions and traces | Session traces for each agent step, with time split between the provider and Cloptima | Request logs with custom properties |
| Spend record and chargeback | Attributed spend record by team, app, environment, and your own fields, with chargeback, AI unit economics, and FOCUS export | Cost tracking with custom properties |
| Telemetry without the gateway | JS, Python, and Go SDKs plus OpenTelemetry GenAI ingest into the same spend model | Logging through its gateway or integrations |