Skip to content

Releases

v0.3.0 — first npm release

Choose a tag to compare

@nxxxsooo nxxxsooo released this 22 Sep 08:06

v0.3.0 — 2026-09-22 (first npm release)

Self-maintained vision router plugin for OpenCode (renamed fork of Chathula/opencode-vision-router@0.2.0, MIT).

What it does

Pasted images are intercepted at the harness level and routed to a cheap vision model whose text output the text-only main agent works from. Multimodal main models skip the extra routing (force: false).

Fixed since upstream 0.2.0

  • V2 capability detection: ctx.model with ctx.catalog.model fallback (fixes TypeError that forced every image through the subagent).
  • Regression tests for the V2 capability path.
  • Quota/rate-limit fallback chain across ordered vision models.

Install

// opencode.jsonc
{
  "plugins": [
    {
      "package": "opencode-multimodal-looker",
      "options": { "model": "alibaba-coding-plan/qwen3.7-plus" }
    }
  ]
}
npm i opencode-multimodal-looker

Directory-path loading still works (see README). Restart opencode after changing config — plugins are not hot-reloaded.

Verification

  • 40 Bun tests, tsc --noEmit, and dist-in-sync CI all pass on the tagged commit.
  • npm view opencode-multimodal-looker version → 0.3.0.

Full diff vs upstream: Chathula/opencode-vision-router@0.2.0...main (fork point) · Repo: https://github.com/nxxxsooo/opencode-multimodal-looker