TECH Signal 465
OpenCode Senses plugin adds 13 local vision tools to text-only OpenCode models
OpenCode Senses is an open-source plugin that attaches a local Moondream vision layer to text-only OpenCode coding models, exposing 13 tools that return OCR, bounding boxes, colors, and other structured evidence from attached images.
Engineers using OpenCode can now drop screenshots and images into a session and have the model reason over them, with inference running entirely on a local GPU and no API keys involved. The cost is a hardware floor of NVIDIA Ampere-or-newer or Apple Silicon with at least 6 GB VRAM, plus a one-time download of Moondream weights and a managed Python runtime. Only the Hacker News feed carried this event, so the signal is a single Show HN post rather than broader press or independent benchmarks, and the project's own framing is the only source for its speed and accuracy claims.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
The plugin is a TypeScript module that runs inside OpenCode and spawns a Python runtime over stdio JSON-RPC, with Moondream 2 as the default model and Moondream 3.1 (9B) supported for larger GPUs.
The 13 senses_* tools (inspect, ocr, detect, point, segment, crop, zoom, colors, diff, annotate, metadata, reverse search, status) return structured, source-grounded evidence, and image content is wrapped in an explicit untrusted-data guard against prompt injection.
Adoption requires `npm install opencode-senses` plus a supported GPU (NVIDIA Ampere or newer, or Apple Silicon M-series) and a first-run setup of a Python venv and Moondream weights; pre-Ampere NVIDIA, AMD, and Intel GPUs are not listed as supported.
THE CLUSTER