Summary

The openai media-understanding provider implements transcribeAudio via transcribeOpenAiCompatibleAudio (Whisper API), but its capabilities array only declared ["image"].

This caused the media-understanding runner to skip the openai provider when processing inbound audio messages (e.g., voice messages on Discord/WhatsApp), resulting in raw audio files being passed directly to agents instead of transcribed text.

Fix

Add "audio" to the openai provider's capabilities array so the runner correctly selects the openai provider for audio transcription when configured with tools.media.audio.

Test

Before fix:

[tools] image failed: Unsupported media type: audio

Agent received raw .ogg file path instead of transcribed text.

After fix:
Audio messages are transcribed via Whisper API before reaching the agent.

Greptile Overview

Greptile Summary

This PR updates the OpenAI media-understanding provider (src/media-understanding/providers/openai/index.ts) to declare support for audio by adding "audio" to its capabilities list. This aligns the provider’s declared capabilities with its existing transcribeAudio implementation, allowing the media-understanding runner to select the OpenAI provider for inbound audio messages when tools.media.audio is enabled, so audio is transcribed before being passed to agents.

Confidence Score: 5/5

• This PR is safe to merge with minimal risk.
• Single-line change that makes declared capabilities match already-exported functionality (transcribeAudio). No behavioral change beyond provider selection logic for audio, and it should unblock intended transcription flow.
• No files require special attention

_{(2/5) Greptile learns from your feedback when you react with thumbs up/down!}