Media & voice (docs/media.md)

gg's media capabilities (image generation, TTS, STT) go through OpenAI-compatible HTTP endpoints:

Capability	Endpoint
Image generation	POST /v1/images/generations
TTS	POST /v1/audio/speech
STT	POST /v1/audio/transcriptions (multipart)

By default the chat provider's baseURL / apiKey are reused (most OpenAI-compatible services offer all three endpoints). Use environment variables for separate configuration:

• GG_MEDIA_BASE_URL / GG_MEDIA_API_KEY
• GG_MEDIA_IMAGE_MODEL / GG_MEDIA_TTS_MODEL / GG_MEDIA_STT_MODEL

Generated files go to ~/.gg/media/{images,tts}/ (0600); STT returns text directly.

Agent tools

• image_generate: prompt → PNG file path. Requires approval (image generation is billed per call).
• tts: text → audio file path (mp3 by default).
• stt: local audio file → transcript text.

CLI

gg media image "一只橘猫在看代码" --size 1024x1024
gg media tts "你好，我是叮当" --format mp3
gg media stt ./voice.m4a --language zh

Limits

• STT audio capped at 64 MiB; image/TTS responses capped at 64 MiB.
• At most 4 images per request.
• Image quality depends entirely on the provider (DALL-E / gpt-image / third-party compatible implementations).