@mizchi/vlmkit-ai

VLM / LLM clients and reasoning pipeline used by VRT to interpret diff images and natural-language intent.

Part of the vrt monorepo.

Install

pnpm add @mizchi/vlmkit-ai

Set OPENROUTER_API_KEY, GEMINI_API_KEY, or ANTHROPIC_API_KEY depending on the provider you target.

Usage

import { buildIntent, runReasoning, askVlm } from "@mizchi/vlmkit-ai";

const intent = await buildIntent({
  baseUrl: "http://localhost:3000",
  description: "Migrate hero section from Tailwind to vanilla CSS",
});

const reasoning = await runReasoning({ /* diff context */ });

What's included

| Module | Purpose | |---|---| | llm-client | Provider-agnostic chat-completion wrapper (Gemini / Anthropic). | | vlm-client | OpenRouter VLM client for diff-image analysis. | | reasoning, reasoning-pipeline | Multi-step structured reasoning over diff context. | | intent | Natural-language intent extraction. | | nlp | Stopword / synonym / heuristic helpers. |

Environment variables

| Variable | Purpose | |---|---| | VRT_LLM_PROVIDER | gemini (default) | anthropic | | VRT_LLM_MODEL | Override provider default model | | VRT_VLM_MODEL | OpenRouter VLM model id | | OPENROUTER_API_KEY | VLM access | | GEMINI_API_KEY | Gemini LLM access | | ANTHROPIC_API_KEY | Anthropic LLM access |

License

MIT