---
title: "Why we bundle every model in the box."
description: "Most AI design tools pick a vendor. Mémoire picks the user. Claude, Codex, Hermes, Ollama, OpenCode, local. Bring your plan, keep your work."
publishDate: 2026-05-10T00:00:00.000Z
author: Sarvesh Chidambaram
tags: ["openness", "models", "harness", "opencode"]
canonical: https://memoire.cv/blog/why-bundle-the-models
---
## TL;DR

Most AI design tools pick a vendor and lock you into their model. Mémoire picks the user. The harness is the product. The model is your choice.

## The vendor lock-in tax

You are paying for a Claude plan. Or a ChatGPT plan. Or both. Maybe you're running Ollama on your machine. Maybe you have an OpenRouter key. Maybe a friend on the team has access to an early-tier OpenCode account.

Every design-AI product on the market wants you to pay them, additionally, for inference you have already paid for once. The math does not work. You should not be paying twice for the same token.

## Bring your own model

Mémoire calls the model you tell it to. It does not host one. It does not resell one. The harness is local-first; the inference call is yours.

Supported runtimes today: **Claude** (via Anthropic API or Claude Code), **Codex** (via OpenAI API or the Codex plugin), **Hermes** (our in-house routing layer), **Ollama** (local), **OpenCode**, **OpenClaw**, and **Cursor** integration. Any one of them. Mix them per task.

## What "harness" means

A harness is the orchestration layer. It owns:

- **Prompt routing**. Which model is best for "design system extraction" vs. "spec generation" vs. "research synthesis"?
- **Tool execution**. Sandboxed shell, file reads, Figma calls, MCP tools.
- **Memory**. The project-memory store every model session pulls from and writes to.
- **Replayable trace**. Every action, every output, every token, saved and inspectable.

The model is one component. Replace it and the rest still works.

## Local-first by default

Mémoire runs on your machine. The CLI, the macOS Studio app, the MCP server. The only network calls are the AI providers you configure and the package registries the CLI installs from. No telemetry. No accounts. No SaaS.

Read the [legal page](/legal) for the short version. Read the [source](https://github.com/sarveshsea/m-moire) for the long one.

## The OpenCode-for-designers manifesto

This is the part the home page section is also about. Don't let one vendor decide how you design. Use the model you have. Move when a better one ships. Take your work with you.

If you want to help build this, [the Notes repo](https://github.com/sarveshsea/memoire-notes) is where the leverage is. Each Note is a graduated workflow that any model can run. The more Notes, the more the harness compounds. The more the harness compounds, the less the vendor matters.