# Bring your own model (BYOM) Bring your own model (BYOM) is the idea that an app, tool, or platform should let **you** decide which AI model it uses, instead of forcing its own model on you, on its own terms. The app provides the interface and the workflow. You provide the brain: a frontier model through your own account, an open-weight model running on your machine, or anything in between. ## The problem it solves Many products bundle AI the same way: one model (or a short list), chosen by the vendor, running on the vendor's servers, billed through the vendor, with usage limits set by the vendor. As a user, you inherit ALL of their decisions: - **Quality**: you're stuck with whatever model they picked, even when a better one exists for your task - **Privacy**: your data goes wherever their model runs, under their data policy - **Cost**: you pay their markup, often hidden in a subscription with quotas - **Continuity**: they can swap, degrade, or remove the model at any time. Your workflow changes without your consent - **Lock-in**: the more your work depends on their AI, the harder it is to leave That's a clear case of [[SaaS Misalignment with User Objectives]]. The vendor's incentives (margins, retention, data) aren't the same as yours. ## What BYOM looks like in practice There are different levels of freedom: 1. **Model picker**: choose among the models the vendor supports. Better than nothing, but still their list 2. **Your own account**: plug in your own API key for a provider. That's [[Bring Your Own Key (BYOK)]]. You pick the provider, and billing and data go through your account 3. **Any compatible endpoint**: point the app to any OpenAI-compatible API, including routers like [[OpenRouter]] 4. **Local models**: run open-weight models on your own hardware with [[Ollama]] or [[LM Studio]]. Nothing leaves your computer BYOK is a subset of BYOM. BYOK is about *whose account* pays. BYOM is about *which model* does the work, wherever it runs. ## Examples - **Code editors and agents**: [[Cursor]] lets you use your own API keys, and tools like [[Cline]] and [[OpenCode]] work with many providers and local models - **Obsidian plugins**: [[Copilot plugin for Obsidian]] and [[Smart Connections plugin for Obsidian]] support multiple providers, including local models (see [[Obsidian AI Plugins]]) - **Terminal tools**: [[Emeraldian]]'s assistant works with Anthropic, OpenAI, OpenRouter, Groq, Ollama, or LM Studio - **My own product**: [[Knowii Voice AI]] runs speech-to-text models locally, and you choose which model to download and use: small and fast, or large and accurate ## Why I care about this I strongly believe that BYOM is the AI version of the [[File over app principle]]. Your notes shouldn't be locked into an app, and your AI capabilities shouldn't be locked into a vendor either. Models improve every few weeks. The best model today won't be the best one in six months. A tool that lets you swap the model keeps getting better over time. A tool that doesn't ages with its vendor's choices. It also matters a lot for [[Data Ownership]]. Some content should never leave your machine (personal journals, customer data, health information). With BYOM, you can use a local model for sensitive work and a frontier model for the rest. ## Trade-offs It's not free, though: - **More setup**: accounts, API keys, endpoints, model downloads - **Inconsistent results**: the app's prompts and features are usually tuned for specific models. A weaker model can make the whole product look bad - **Hardware requirements** for local models - **Support burden** for the vendor, which is why many of them avoid it - **Your responsibility**: you manage the costs, rate limits, and security of your keys When evaluating an AI tool, ask yourself: can I change the model? If the answer is no, you're renting the vendor's judgment along with the tool. ## References - ## Related - [[Bring Your Own Key (BYOK)]] - [[SaaS Misalignment with User Objectives]] - [[File over app principle]] - [[Data Ownership]] - [[Local-first software]] - [[Self-hosting]] - [[Large Language Models (LLMs)]] - [[Small Language Models (SLMs)]] - [[Ollama]] - [[LM Studio]] - [[OpenRouter]] - [[Obsidian AI Plugins]] - [[Knowii Voice AI]]