Make an image from a request or references
Describe an illustration or provide images to edit or combine. The agent uses the configured image resource, saves the result to a writable destination and can attach it to a conversation or document.
References can come from attachments, documents, email, the console or the web. Requested dimensions are matched to the model’s native sizes and actual dimensions are returned. Editing, reference limits and formats depend on the selected resource.
Turn a recording into meeting notes
Transcription accepts audio or video resources. Long recordings are segmented; the transcript is retained as a file and a hierarchical summary brings the content together. Meeting and video summarisation instructions are configurable.
For a public YouTube URL, Galaris retrieves available captions. It does not download the video or fabricate a transcript when captions are missing. Readers can return to the transcript to check a decision or quotation.
Understand more than speech
Vision analyses an image in response to a question. Audio and video functions can examine ambience, instruments, audible events or video content. These use specialised resources distinct from speech transcription.
Multimedia reading is bounded to 32 MB and 20 minutes in the inspected implementation. The model must support the requested input.
Track long generations
| Output | Integrations for this workflow |
|---|---|
| Sound effects | ElevenLabs, SunoAPI.org sound service |
| Music | Eleven Music, Lyria through OpenRouter, SunoAPI.org |
| Video | OpenRouter, Seedance through BytePlus LAS |
A request provides instructions, supported options and a destination. It returns a Process reference so the conversation can continue. Success is recorded after the files are written and linked to the originating Task.
Recover the result without generating it again
Unsupported options are rejected. Invocation identity separates retrying the same request from requesting a new variation. An uncertain submission or write can require an explicit resolution.
Delivery repair can attach an existing file or permit another write after verification. It does not automatically purchase another generation. Unknown costs remain labelled unknown.
The Multimedia connection is active on initial setup; an existing decision to disable it is preserved. Enable the required resources and check the service account; SunoAPI.org is a third-party service with a separate account. Galaris does not automatically expose every commercial feature of each provider.
Prepare meeting notes · Work with files
Keep a reusable image description
image_read saves its description before reporting success. For an active document attachment, it enriches the memory companion with revision and provenance. For another URI, it remains private to the agent/resource pair. A storage failure fails the tool; the description does not share the source file.
Inspect attachments directly in the conversation
With the internal engine and a compatible model, images, audio, video and PDFs attached to messages can reach the model directly, including retained history. The agent can inspect the media without always calling a separate analysis tool.
Support depends on the configured model, format and transport. Agent access rights and a total byte budget are checked; current messages take priority. A link mentioned in text does not trigger a download on its own. Dedicated tools remain useful for lasting transcripts, saved descriptions or unsupported formats. Dream video analysis still uses only the audio track.