Model Governance
Model governance means knowing which models your agents use, allowing only the ones you approve, controlling who can change prompts and model settings, and keeping a record. This page maps those needs to ANTS Platform features.
ℹ
There is no llmops.governance SDK, policy engine API or model approval workflow API. Governance is configured in the app. The features below are where each part lives.
Which models are in use
| Need | Feature |
|---|---|
| See every model your agents call | Generations in Tracing record the model name. The Cost Dashboard and Latency Dashboard break cost and latency down by model. |
| See models across the organization | AI Inventory lists LLMs and agent frameworks with lifecycle status and end-of-life dates. |
| Know when a model reaches end of life | The catalog's current, deprecated and archived status and EOL dates. See Model Lifecycle. |
Allow only approved models
- Approvals. Approve models and frameworks per organization in AI Inventory. ANTS Platform flags any model in your traces that is not approved, in the AI Agent Center and on each agent's Risk tab. See Model Lifecycle.
- Enforce at the gateway. With Budget AI, applications call model groups instead of provider models, and each key is limited to named model groups. A request for any other model is refused. See Model Groups & Routing and Keys.
- Guardrails. AI Guardrails scan and can block or redact content in agent traffic.
Control who can change things
- Roles. Access to prompts, datasets, evaluators, LLM connections and Budget AI is set by role. See Members & Roles and Custom Roles.
- Protected prompt labels. Protect the
productionlabel so only permitted users can change which prompt version is live. See Prompt Management. - Provider credentials. LLM connections and Budget AI credentials are stored encrypted and are visible only as masked values.
Check quality before and after a change
- Run experiments on a dataset and score them with LLM-as-a-Judge before you switch a model or prompt.
- Keep evaluators running on live traces to watch quality after the change.
- The AI Optimizer tests cheaper or faster models against an agent's real traces and recommends a switch only when quality holds.
ANTS Platform records evaluation results. It does not block a deployment based on scores; make that decision in your own CI job.
Keep a record
- Audit Logs record changes to prompts, automations, LLM connections and other project resources.
- Governance evidence and framework coverage live under Governance in the sidebar. See AI Governance.
Next steps
Planned capabilities are listed on the roadmap.