RFC-0005: Attachments & Multi-modality v1.0

Status: Proposed
Created: 2026-02-01
Authors: OpenIntent Contributors
Requires: RFC-0001 (Intents)

---

Abstract

This RFC defines support for file attachments on intents, enabling multi-modal workflows involving images, audio, video, documents, and other binary content.

Motivation

Modern AI agents frequently work with multi-modal content:

• Vision tasks: Image analysis, OCR, visual QA
• Audio processing: Transcription, voice commands, music analysis
• Document workflows: PDF parsing, contract review, data extraction
• Video understanding: Content moderation, scene detection, summarization

The protocol must provide a standard mechanism for associating binary content with intents so that agents can exchange rich media as part of structured coordination.

Data Model

Attachment Object

{
  "id": "uuid",
  "intent_id": "uuid",
  "filename": "document.pdf",
  "mime_type": "application/pdf",
  "size": 1048576,
  "storage_url": "https://storage.example.com/files/abc123",
  "metadata": {
    "width": null,
    "height": null,
    "duration": null,
    "pages": 24
  },
  "uploaded_by": "agent-id",
  "created_at": "ISO 8601"
}

Supported Metadata

Metadata fields are content-type specific:

Field	Applies To	Description
width	Images, Video	Width in pixels
height	Images, Video	Height in pixels
duration	Audio, Video	Duration in seconds
pages	Documents	Number of pages
encoding	Audio	Audio encoding format
frame_rate	Video	Frames per second

Endpoints

Method	Path	Description
POST	/v1/intents/{id}/attachments	Add attachment
GET	/v1/intents/{id}/attachments	List attachments
DELETE	/v1/intents/{id}/attachments/{attachmentId}	Remove attachment

Example: Image Analysis Workflow

# Upload an attachment reference
curl -X POST http://localhost:8000/api/v1/intents/{id}/attachments \
  -H "X-API-Key: dev-user-key" \
  -d '{
    "filename": "receipt.jpg",
    "mime_type": "image/jpeg",
    "size": 245000,
    "storage_url": "https://storage.example.com/receipts/r123.jpg",
    "metadata": { "width": 1920, "height": 1080 }
  }'

Storage Considerations

This RFC defines the attachment metadata model only. The actual file storage mechanism (S3, GCS, local filesystem, etc.) is left to the implementation. The storage_url field provides the indirection needed to support any storage backend.

Integration with Other RFCs

• RFC-0015 (Agent Memory): Memory entries can reference attachments for rich context
• RFC-0008 (LLM Integration): Adapters can include attachment content in LLM prompts for multi-modal inference