> Markdown twin of https://conduitllm.com/use-cases/bring-your-own-key.html > Use your own Anthropic, OpenAI, or Gemini API key in a free desktop AI client. Seventeen providers, no markup, no subscription — keys stay in your OS keychain. --- 1. [Conduit](https://conduitllm.com/) 2. [Use cases](https://conduitllm.com/use-cases.html) 3. Bring your own key Bring your own key # Pay *per token*, not per seat Use the API keys you already have — Anthropic, OpenAI, Gemini, xAI, DeepSeek, OpenRouter, and eleven more providers — in one desktop app. You pay each provider's own per-token price. Conduit adds nothing on top, and never resells a token. [Download Conduit](https://conduitllm.com/#get) [Is this cheaper for you?](https://conduitllm.com/api-vs-subscription.html) - Seventeen providers, one app - Keys live in your OS keychain - No subscription, no markup The problem You already pay for Claude. And ChatGPT, for the days Claude is slow. And Gemini, because it came with something else you use. Three subscriptions, most of every seat unused, and still no single window that shows all three models side by side. What bringing your own key gets you ## One app, every provider's own price No seat, no bundle, no vendor deciding which models you're allowed to compare. ### No markup, no seat Your request goes straight to the provider you picked, at that provider's own per-token price. Conduit doesn't take a cut, doesn't bundle a plan, and doesn't charge you at all — the app itself is free. ### Your key never enters the interface Keys are held in your OS keychain — Windows Credential Manager, macOS Keychain, or Linux Secret Service — and referenced by name. The screen you type into can request a completion; it cannot read a key or open a socket. ### Switch providers with a keystroke Paste a key once and ⌘⇧P cycles the active provider mid-conversation. Compare a frontier model against a cheap fast one on the same question without opening a second app. How it works ## Paste a key, pick a model No account with Conduit to create, because there isn't one. ### Open Providers & Keys ⌘, opens Settings. Choose a provider from the list and paste the API key you generated in that provider's own console. ### Pick a model from the list Conduit fetches Conduit calls the provider's own model-listing endpoint rather than shipping a hardcoded catalogue, so what you see is whatever that provider currently offers your key. ### Add another provider, or a gateway, any time Repeat with a second or third provider, or paste a single OpenRouter or OpenCode Zen key to reach many families at once. ⌘⇧P switches which one answers next. Setup, costs and limits for each provider are in the [provider guides](https://conduitllm.com/providers.html), starting with [OpenRouter](https://conduitllm.com/providers/openrouter.html). Three ways to bring a key ## Whichever way you reach a model, Conduit talks to it the same way Your credential, your request, your bill — Conduit is the same client on all three. ### A named adapter Anthropic, OpenAI, Google Gemini, Groq, DeepSeek, Mistral, xAI, Z.ai, Moonshot AI, Qwen, Together AI, and Fireworks AI each have a dedicated provider. Paste the key, load the model list, done. ### A gateway, one key OpenRouter or OpenCode Zen sit behind a single key and cover model families with no dedicated adapter of their own — MiniMax is reachable this way, not through a MiniMax provider, because Conduit doesn't have one. ### A custom endpoint, or none at all Eleven of the seventeen providers, including Anthropic and OpenAI, accept an editable base URL, so a self-hosted proxy such as LiteLLM works the same way a paste-a-key setup does. Ollama and LM Studio need no key or internet at all. Conduit also charts estimated cost and token totals per day, computed on your own machine from a small, hand-maintained price table — useful for noticing a trend, not a substitute for your provider's invoice. Limits ## What it does, and what it doesn't Bringing your own key moves work onto you that a subscription hides. Here's exactly what. ### What it does - Connects to seventeen providers with your own key — the OpenAI-compatible endpoint is the seventeenth, not an extra - Sends requests straight from your machine to the provider, with no Conduit server in between - Stores every key in your OS keychain, never in the interface or in Conduit's settings as a value - Accepts a custom base URL on eleven providers, so a self-hosted proxy like LiteLLM works - Reaches gateway-only families like MiniMax through OpenRouter or OpenCode Zen - Charts estimated per-day cost and token totals from your own usage, computed locally ### What it doesn't - Doesn't create an account with any provider for you, or handle billing — you sign up and pay each one yourself, or pay one gateway - Doesn't guarantee the API is cheaper for you — heavy users on a flat plan can pay more per token; [check the calculator](https://conduitllm.com/api-vs-subscription.html) - Doesn't sync anything across devices — there's no Conduit account, so no cloud sync between machines - Doesn't offer a dedicated MiniMax provider, or an Azure OpenAI or AWS Bedrock provider - Doesn't report your invoice — cost figures are estimates from a small hand-maintained price table, not billing data - Doesn't offer the provider's own hosted web search through a gateway or custom base URL — only on the official Anthropic, OpenAI, and Gemini endpoints and OpenRouter (local search backends, Exa by default, work anywhere) ## Bring-your-own-key questions **What does "bring your own key" mean?** You create an account with an AI provider — Anthropic, OpenAI, Google, or another — generate an API key from their console, and paste it into Conduit yourself. Conduit is the client, not the vendor: it never issues you a key, never proxies your requests through a server of its own, and never adds a markup on top of what the provider charges. The relationship, the invoice, and the credential are between you and the provider. **Which providers can I use with my own API key?** Seventeen are built in: Anthropic, OpenAI, Google Gemini, OpenRouter, OpenCode Zen, Groq, DeepSeek, Mistral, xAI, Z.ai, Moonshot AI, Qwen, Together AI, Fireworks AI, LM Studio, Ollama, and any OpenAI-compatible endpoint — that last one is the seventeenth, not an addition on top of the other sixteen. Ollama and LM Studio run locally and need no key at all. **Is paying per token cheaper than a subscription like ChatGPT Plus or Claude Pro?** It depends on how much you send and how long your conversations run — the whole history is re-sent every turn, so long chats cost more than short ones. [A calculator and worked examples](https://conduitllm.com/api-vs-subscription.html) walk through the break-even point. If you lean on one model heavily every day, a flat-rate subscription can still be the better deal. **Can I use a gateway like OpenRouter instead of a key per provider?** Yes. OpenRouter and OpenCode Zen sit behind a single key and give you a fetched, changing catalogue of models, including families with no dedicated adapter of their own — MiniMax is the clearest example. Conduit reads whichever list the gateway returns rather than a fixed catalogue baked into the app. **Can I point Conduit at my own proxy, like LiteLLM?** Yes, on eleven of the seventeen providers, including Anthropic, OpenAI, and the generic OpenAI-compatible provider. Each of those has an editable base URL field, so a self-hosted proxy such as LiteLLM or vLLM, or a regional mirror of a provider's API, works the same way a paste-a-key setup does. The other six — Gemini, OpenRouter, OpenCode Zen, Groq, DeepSeek, and Mistral — use a fixed endpoint. **Where does Conduit store my API keys?** In your operating system's own credential store — Windows Credential Manager, macOS Keychain, or Linux Secret Service. Conduit's settings screen holds a reference to the entry, never the key value, and the interface layer that renders your chat never has access to the secret itself. Requests go straight from your machine to the provider you picked; there is no Conduit server in between. Related ## Other ways people use Conduit For local-model users ### Local models, with a real app around them Ollama, LM Studio, or your own server — no key, no bill, one keystroke from a cloud model when you want one. [Run models locally](https://conduitllm.com/use-cases/local-llm.html) For sensitive work ### An AI assistant that doesn't phone home No telemetry, no account, local storage — and fully offline with a local model. Keep it private For developers ### An MCP client that asks first Local and remote servers, OAuth, registry install, and consent graded by what a tool can do. Conduit as your MCP client Every feature, with its limits, is on [the full feature list](https://conduitllm.com/features.html); how keys and every other network flow are handled on [the privacy page](https://conduitllm.com/privacy.html), and provider setup steps in [the documentation](https://conduitllm.com/docs.html). ## Bring the key you already have Free, open source, and yours to run — no account, no trial, nothing to cancel. [Download Conduit](https://conduitllm.com/#get) [Read the setup docs](https://conduitllm.com/docs.html) A release candidate: installers aren't OS code-signed yet, so expect a SmartScreen or Gatekeeper prompt on first launch.