@tekmemo/benchmark-kit

Provider-neutral benchmarking toolkit for TekMemo packages and adapters.

This package helps benchmark:

• memory stores
• embedder adapters
• recall stores
• rerankers
• package-level workflows
• future provider adapters

It does not own production product behavior. It only measures it.

Install

pnpm add -D @tekmemo/benchmark-kit

Basic usage

import {
  BenchmarkRunner,
  createBenchmarkSuite,
  markdownBenchmarkReport
} from "@tekmemo/benchmark-kit";

const suite = createBenchmarkSuite({
  name: "local-memory",
  cases: [
    {
      name: "write-core-memory",
      iterations: 100,
      warmupIterations: 10,
      async run() {
        await store.write("memory/core.md", "# Core Memory");
      }
    }
  ]
});

const runner = new BenchmarkRunner();
const result = await runner.runSuite(suite);

console.log(markdownBenchmarkReport(result));

Recall benchmark

import { createRecallQueryBenchmarkCase } from "@tekmemo/benchmark-kit";

const benchmarkCase = createRecallQueryBenchmarkCase({
  name: "upstash-query-top-10",
  store,
  query,
  iterations: 100,
  warmupIterations: 10
});

Thresholds

import { evaluateBenchmarkThresholds } from "@tekmemo/benchmark-kit";

const verdict = evaluateBenchmarkThresholds(result, {
  maxP95Ms: 200,
  maxErrorRate: 0.01
});

Package boundary

This package owns:

• benchmark runner
• latency measurement
• throughput calculation
• error-rate calculation
• threshold evaluation
• JSON/Markdown reporters
• fake targets for tests
• provider-neutral benchmark helpers

It does not own:

• .tekmemo/ protocol
• vector recall implementation
• embeddings implementation
• reranking implementation
• cloud quotas
• billing