@tekmemo/openai

@tekmemo/openai

Production-ready OpenAI embedding adapter for TekMemo.

This package is intentionally provider-specific and BYOK-ready. It accepts an OpenAI API key from the host application and does not store secrets.

Install

pnpm add @tekmemo/openai

@tekmemo/openai uses the official openai SDK for production API calls and keeps a small injectable embeddings-client interface for tests and custom hosts.

Quickstart

import { createOpenAIEmbedder } from "@tekmemo/openai";

const embedder = createOpenAIEmbedder({
  apiKey: process.env.OPENAI_API_KEY!,
  model: "text-embedding-3-small",
  dimensions: 1024
});

const result = await embedder.embedTexts({
  texts: ["TekMemo is file-first memory infrastructure."]
});

Recommended TekMemo defaults

For most TekMemo recall flows:

createOpenAIEmbedder({
  apiKey: process.env.OPENAI_API_KEY!,
  model: "text-embedding-3-small",
  dimensions: 1024
});

OpenAI's embedding guide says text-embedding-3-small defaults to 1536 dimensions and text-embedding-3-large defaults to 3072 dimensions, and the dimensions parameter can reduce those vector sizes for compatible v3 embedding models.

BYOK

createOpenAIEmbedder({
  apiKey: userProvidedOpenAIKey,
  model: "text-embedding-3-small"
});

The package does not persist, encrypt, or log keys. TekMemo Cloud BYOK storage belongs in closed-source cloud code.

Production features

• official OpenAI SDK client for embeddings
• BYOK-ready config
• model and dimension validation
• batch splitting
• SDK retry/timeout support
• response shape validation
• embedding count validation
• finite-number vector validation
• fake OpenAI client for unit tests

What this package does not own

• .tekmemo/ file protocol
• vector storage
• recall store contracts
• reranking
• billing
• cloud BYOK encryption
• tenant/provider routing