Vector adapters

Vector adapters implement the RecallStore contract, allowing TekMemo to store and search memory using semantic vector search.

Upstash Vector

@tekmemo/upstash-vector provides a serverless recall store backed by Upstash's global vector database. It is optimized for low-latency retrieval in serverless environments like Cloudflare Workers and Vercel.

Install

npm install @tekmemo/upstash-vector

Usage

import { createUpstashRecallStore } from "@tekmemo/upstash-vector";

const store = createUpstashRecallStore({
  url: process.env.UPSTASH_VECTOR_REST_URL,
  token: process.env.UPSTASH_VECTOR_REST_TOKEN,
  namespace: "my-project-production" // Optional
});

// Search for relevant memory
const results = await store.query({
  query: "How do I configure the CLI?",
  limit: 5,
  includeMetadata: true
});

console.log(`Found ${results.items.length} relevant fragments.`);

Features

• Namespace Isolation: Supports multi-tenant or multi-project setups using Upstash namespaces.
• Filter Builder: Provides a structured DSL for filtering recall results by tags, kinds, or custom metadata.
• Metadata Normalization: Automatically handles the conversion between TekMemo's memory model and Upstash's vector metadata format.
• Zero-Dependency Transport: Uses standard fetch for communication, making it compatible with all modern JavaScript runtimes.

Use when

Use the Upstash adapter when you need a hosted, zero-maintenance vector store for project memory that scales with your application.