Roder
Capabilities

Media generation

Roder generates images through a provider-neutral media API. Tool calls, app-server clients, SDKs, and the TUI all use the same request shape, and every output is persisted as a Roder-owned MediaArtifact before it is shown or attached to another turn.

Image providers

  • openai supports gpt-image-2 plus compatibility GPT Image model ids.
  • google supports Gemini Nano Banana image models, including aspect ratio and image-size options where available.
  • fake is deterministic and always available for tests and offline profiles.
[media]
artifacts_dir = "~/.roder/artifacts"

[media.image_generation]
default_provider = "openai"
default_model = "gpt-image-2"
max_outputs = 4
max_input_images = 16

CLI and app server

roder media providers --json
roder media models --provider google
roder media generate "a clean terminal product hero" --provider openai --model gpt-image-2

App-server clients call media/image/providers/list andmedia/image/generate. Generated artifacts then flow through the existingmedia/list, media/read, media/thumbnail,media/delete, and media/attachToTurn methods.