@prompt2md/core

Dual-engine processing pipeline: content sniffer, engine router with evidence-based escalation, LiteLLM gateway factory, Markdown IR, and token reporting. See ADR-001 and ADR-002.

Usage

import {
  convertDocument,
  createDoclingEngine,
  createLiteLlmGateway,
  createMarkitdownEngine,
  createPromptOptimizerEngine,
} from "@prompt2md/core";

const gateway = createLiteLlmGateway({
  baseUrl: "http://localhost:4000/v1",   // LiteLLM proxy
  defaultModel: "claude-sonnet-5",
  fallbackModels: ["gpt-4.1"],
});

const deps = {
  engines: {
    "prompt-optimizer": createPromptOptimizerEngine(gateway),
    markitdown: createMarkitdownEngine(),                          // needs: pip install "markitdown[all]"
    docling: createDoclingEngine({ baseUrl: "http://localhost:5001" }), // docling-serve container
  },
};

const outcome = await convertDocument({ kind: "file", path: "./report.pdf" }, deps, {
  tokenBudget: 6000,
});
console.log(outcome.report);   // TokenReport: input/output tokens, ratio, per-section
console.log(outcome.markdown);

Engine sidecars

| Engine | Runtime | Setup | |---|---|---| | markitdown | persistent Python subprocess (python/markitdown_worker.py) | pip install "markitdown[all]" | | docling | docling-serve REST | docker run -p 5001:5001 quay.io/docling-project/docling-serve (use the -cu* image + docling-tools models download in production) | | prompt-optimizer | LiteLLM proxy | pip install litellm[proxy] && litellm --port 4000 |

Tests

pnpm --filter @prompt2md/core test        # 44 tests: golden-corpus routing conformance + unit
pnpm --filter @prompt2md/core typecheck