@nfinitmonkeys/llm-cortex

Cortex API adapter for @nfinitmonkeys/brain. Implements the LLM interface natively for chat + embeddings; synthesizes structuredExtract + summarize via JSON-mode chat.

Install

npm install @nfinitmonkeys/llm-cortex @nfinitmonkeys/brain

Quick start

import { Brain } from '@nfinitmonkeys/brain';
import { CortexLLM } from '@nfinitmonkeys/llm-cortex';

const llm = new CortexLLM({
  apiKey: process.env.CORTEX_API_KEY!,
  // baseURL defaults to https://cortexapi.nfinitmonkeys.com/v1
  // defaultModel defaults to Qwen/Qwen2.5-72B-Instruct-AWQ
  // defaultEmbedModel defaults to BAAI/bge-m3
});

const brain = new Brain({ vaultRoot: '~/Projects/my-clan/.brain', llm });

// Now LLM-driven brain operations work
const answer = await brain.query('what was the EV market hallucination?');

Capability map

| Operation | Implementation | |---|---| | chat | OpenAI-compatible /v1/chat/completions via openai SDK | | embed | OpenAI-compatible /v1/embeddings via fetch (default model: BAAI/bge-m3, 1024-dim) | | structuredExtract | JSON-mode chat with the schema embedded in the system prompt | | summarize | Tuned chat prompt with a target word count | | iris | Not implemented — Cortex's Iris endpoint isn't yet exposed via this adapter. Will land in a later version when the Cortex API surface is stable. | | vlm / tts / stt | Not implemented yet — same reason. |

For LLM operations not natively implemented, callers can use embedOrFallback, structuredExtractOrFallback, summarizeOrFallback from @nfinitmonkeys/brain which synthesize via chat.

Environment defaults

The CortexLLM constructor accepts apiKey directly, but you may also use environment variables:

• CORTEX_API_KEY (required if not passed in)
• CORTEX_BASE_URL (optional; defaults to https://cortexapi.nfinitmonkeys.com/v1)
• CORTEX_DEFAULT_MODEL (optional)
• CORTEX_DEFAULT_EMBED_MODEL (optional; defaults to BAAI/bge-m3)

const llm = CortexLLM.fromEnv();  // pulls all four from process.env

License

MIT.