@restormel/reasoning-core

Knowledge Verify pipeline — claim extraction, reasoning quality evaluation, and epistemic constitution checks. Host apps inject LLM routing and generateText via ReasoningCoreContext (no SOPHIA vertex, DB, or engine imports).

Install

npm install @restormel/reasoning-core @restormel/contracts

Peer: ai (Vercel AI SDK) when wiring real models in the host adapter.

Usage

import {
  runVerificationPipeline,
  type ReasoningCoreContext,
} from "@restormel/reasoning-core";

const ctx: ReasoningCoreContext = {
  generateText: (params) => hostGenerateText(params),
  resolveExtractionRoute: (opts) => hostResolveExtraction(opts),
  resolveReasoningRoute: (opts) => hostResolveReasoning(opts),
  trackTokens: (input, output) => hostTrackTokens(input, output),
};

const result = await runVerificationPipeline(
  { text: "Argument to verify…" },
  {
    ctx,
    includePassOutputs: false,
  }
);

When includePassOutputs is true, pass a host hook (e.g. SOPHIA runDomainAgnosticReasoning):

await runVerificationPipeline(request, {
  ctx,
  runPassOutputs: (inputText, callbacks, opts) =>
    runDomainAgnosticReasoning(inputText, callbacks, opts),
});

Publish

Tag platform-v* in the restormel-keys monorepo — see docs/restormel/SUITE-ARCHITECTURE-MIGRATION.md Phase 3.

Related

• Types/schemas: @restormel/contracts/verification, @restormel/contracts/constitution
• SOPHIA consumer adapter: src/lib/server/verification/sophiaAdapter.ts in the sophia repo