Add the BEACON DESIGN scaffold for wiring a ComfyUI IMAGE into the existing
generic ChatCompletion node so a vision-capable model can describe/understand
images on either backend.
- specs/009-vlm-image-input/spec.md — WHAT/WHY: optional IMAGE input, same
node on both backends, structured-output-with-image, graceful degradation
on non-vision models; text-only path unchanged when no image is wired
- ADR-008 (Proposed) — carry images on an optional Message.images field; each
provider translates to its own wire shape (Ollama /api/chat images,
llama.cpp OpenAI image_url parts, shared chat_structured multimodal
content). Extends ADR-007's adapter pattern to a second input modality
- epics/vlm-image-input.md — new epic with success criteria/non-goals; spec
backlinked via .beacon.toml and listed under the epic
- Roadmap + ADR index updated
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UvS9TFMCFYNHC4MMvzJaZS
Claude
45e49a98eb