# Omnia vs Foundry Agent Service

> Omnia rolls out each version in steps with automatic rollback, tests voice agents end to end, and deploys the same pack to Foundry or your own cluster. In Foundry each endpoint serves one version, and simulated-conversation evaluation doesn’t cover voice.

Category: Cloud agent platforms. Checked against Microsoft’s published material on 2026-10-22. Canonical page: https://www.altairalabs.ai/vs/azure-ai-foundry/

## Foundry runs agents inside Microsoft’s stack, and Omnia can deploy to it with a stepped rollout and voice tests added.

Omnia moves traffic to a new version in steps, pulls it back on the metrics, and tests voice agents as calls. Foundry Agent Service publishes into Teams with an Entra identity, but each endpoint serves one version, so a release reaches everyone the moment it’s live.

## When it matters: A voice agent on the customer line gets a new version.

**On Omnia:** Simulated callers have already rung it, interrupting, talking over it and leaving long silences, and it passed. It goes live on a share of calls first. On real lines its interruption handling is worse, the metrics show it, and the previous version is back before the morning peak. (Omnia docs: https://promptarena.altairalabs.ai/arena/how-to/voice/setup-voice-testing/)

**On Foundry Agent Service:** An agent endpoint serves one version at a time and routes all of its traffic to it; traffic splitting between versions isn’t supported. Simulated-conversation evaluation isn’t supported for voice agents. (source: https://learn.microsoft.com/en-us/azure/foundry/agents/concepts/hosted-agents)

## A slice of traffic first

Old and new versions serve at the same time. A call that started on the old one finishes there, and the new one gets more calls as its metrics hold. If they don’t, it’s rolled back. *(enterprise)* (Omnia docs: https://omnia.altairalabs.ai/tutorials/progressive-rollouts/)

Foundry Agent Service: One version per endpoint, all traffic. Roll back by pinning an earlier version. (source: https://learn.microsoft.com/en-us/azure/foundry/agents/concepts/hosted-agents)

## Voice agents get tested too

Voice scenarios run full duplex against the agent, and each session is recorded so you can listen back. (Omnia docs: https://promptarena.altairalabs.ai/arena/how-to/voice/setup-voice-testing/)

Foundry Agent Service: Single-turn and simulated-conversation evaluation aren’t supported for voice agents. The red-teaming agent supports text scenarios only. (source: https://learn.microsoft.com/en-us/azure/foundry/agents/overview)

## Rules on tool calls

A CEL rule reads each tool call’s arguments and the caller’s identity, and allows or denies it deterministically. If the broker can’t answer, the call is refused. *(enterprise)* (Omnia docs: https://omnia.altairalabs.ai/how-to/security/configure-tool-policies/)

Foundry Agent Service: Tool-call guardrails, in preview, run content-risk classifiers on arguments for a listed set of tools. (source: https://learn.microsoft.com/en-us/azure/ai-foundry/guardrails/intervention-points?view=foundry)

## At a glance

| Topic | Omnia | Omnia docs | Foundry Agent Service | Foundry Agent Service docs | Source |
|---|---|---|---|---|---|
| Immutable versions | A versioned pack: prompts, tools, policy and tests. | https://omnia.altairalabs.ai/reference/core/promptpacksource/ | Immutable agent versions. | documented | https://learn.microsoft.com/en-us/azure/foundry/agents/overview |
| End-user identity to tools | OIDC or edge trust, carried to every call. | https://omnia.altairalabs.ai/how-to/security/configure-authentication/ | Entra Agent ID and on-behalf-of. | documented | https://learn.microsoft.com/en-us/azure/foundry/agents/concepts/hosted-agents |
| Traffic splitting | Stepped rollout with automatic rollback. | https://omnia.altairalabs.ai/tutorials/progressive-rollouts/ | Not supported. | not documented | https://learn.microsoft.com/en-us/azure/foundry/agents/concepts/hosted-agents |
| Voice agent testing | Simulated callers, barge-in, dead air. | https://promptarena.altairalabs.ai/arena/how-to/voice/setup-voice-testing/ | Not supported. | not documented | https://learn.microsoft.com/en-us/azure/foundry/agents/overview |
| Adversarial testing | Against the agent’s real tools. | https://promptarena.altairalabs.ai/arena/how-to/voice/voice-red-team/ | Red-teaming agent, text only, mock tools. | partly documented | https://learn.microsoft.com/en-us/azure/foundry/concepts/ai-red-teaming-agent |
| Deterministic argument policy | A CEL rule on every call. *(enterprise)* | https://omnia.altairalabs.ai/how-to/security/configure-tool-policies/ | Classifier guardrails, preview. | partly documented | https://learn.microsoft.com/en-us/azure/ai-foundry/guardrails/intervention-points?view=foundry |
| Runs outside Azure | Your cluster, AWS or Google. | https://promptarena.altairalabs.ai/arena/explanation/deploy/overview/ | Not documented. | not documented | https://learn.microsoft.com/en-us/azure/foundry/agents/concepts/hosted-agents |

## Questions

**Can I use Omnia with Foundry?**

Yes. Omnia deploys a pack to Azure AI Foundry with one command, uses Azure OpenAI and Foundry models, and runs its scenarios against the deployed agent.

**Does Omnia publish agents to Teams?**

Omnia serves agents over WebSocket, REST, A2A or MCP. Deployed to Foundry, the same agent can use Foundry’s publishing.