chattydeer

A Node.js minimal chat completions tool

A Node.js minimal LLM chat completionist built on top of @jsilvanus/embedeer.

Provides Explainer, LLMAdapter, and prompt utilities for structured LLM-based explanation and narration tasks.

Install

npm install @jsilvanus/chattydeer

Usage

import { Explainer } from '@jsilvanus/chattydeer';

const explainer = await Explainer.create('llama-3.2-3b', { deterministic: true });

const result = await explainer.explain({
  task: 'narrate',
  domain: 'evolution',
  context: { filePath: 'src/auth/handler.ts' },
  evidence: [
    { id: 1, source: 'src/auth/handler.ts', excerpt: '2024-03-15 *** LARGE CHANGE' },
  ],
  maxTokens: 256,
});

console.log(result.explanation); // "The auth handler underwent a major rewrite..."
await explainer.destroy();

API

• Explainer.create(modelName, opts) — create an explainer bound to a model
• explainer.explain(request) — explain using structured evidence blocks
• LLMAdapter.create(modelName, opts) — low-level text-generation adapter
• explainForGitsema(payload, opts) — gitsema-compatible adapter
• renderTemplate(domain, vars) — render a domain-specific prompt template
• estimateTokensFromChars(chars) / trimEvidenceForBudget(prelude, evidence, budget) — prompt utilities

See explainer-contract.md for the full interface contract.

License

MIT