Multi Modal Input Handling With Attachments And Fragments

via “multi-modal input handling with attachments and fragments”

CLI tool for interacting with LLMs.

Unique: Provides a unified Attachment abstraction that handles format conversion and provider-specific encoding automatically, allowing the same code to work with different vision models. Fragments allow inline references to attachments in prompts, enabling natural multi-modal interactions.

vs others: More transparent than manually encoding images to base64 because attachment handling is automatic; more flexible than model-specific vision APIs because it abstracts provider differences; simpler than building custom multi-modal pipelines because attachments are first-class in the Prompt API.

via “multi-modal input handling with attachments and fragments”

CLI for LLMs — multi-provider, conversation history, templates, embeddings, plugin ecosystem.

Unique: Uses a Fragments abstraction to represent different media types uniformly, allowing the same Prompt class to handle text, images, audio, and files without conditional logic. Attachments are persisted to the conversation log, making multi-modal conversation history queryable and reproducible.

vs others: More unified than OpenAI's API because it abstracts away provider-specific attachment formats, and more persistent than Anthropic's approach because attachments are logged to the database for future reference.