Back to catalog / Agent workflows

Build a routed multimodal in-app voice assistant

Route knowledge, account, and web-search requests to specialist agents, then expose the same workflow through text and voice.

Goal

Route knowledge, account, and web-search requests to specialist agents, then expose the same workflow through text and voice.

Define separate knowledge, synthetic account-lookup, and web-search agents. [src-openai-build-routed-multimodal-inapp-voice-assistant]
Route by intent through triage/handoffs and inspect traces for the selected tool path. [src-openai-build-routed-multimodal-inapp-voice-assistant]
Add audio input/output and spoken style, then compare factual consistency across text and voice. [src-openai-build-routed-multimodal-inapp-voice-assistant]

Knowledge and account requests must reach different agents; text and voice answers must agree on facts for the same test request.

If routing is wrong, refine handoff descriptions and add ambiguous cases; if speech differs from text, inspect transcription and tool traces.

OpenAI Agents SDK and API key, approved knowledge files, synthetic account data; audio dependencies for voice mode.

Sends requests/audio to model and speech APIs; account tools are read-only with synthetic users, and API calls may be billed.

Example routing does not replace production authentication, tenant isolation, or access control.

These values describe the upstream repository, not popularity or likes for an individual skill or playbook.

No metrics snapshot is available yet.

License: MIT