How to run an AI coding agent without your own API key

Every "free" AI coding agent eventually hits the same wall: paste an API key to continue. Liminal AI removes that wall two ways — bring your own key and pay your provider directly, or use managed inference and skip key setup entirely. Here's how each works, and when to pick which.

The API-key tax on getting started#

An AI coding agent is only useful once it can call a model. For most local and open-source agents that means a setup detour before you've written a single prompt: sign up for OpenAI or OpenRouter, add a payment method, generate a key, find the right base URL, and paste it into a config file. It's not hard — it's just friction, and it's enough to make a lot of people bounce before they see the tool actually work.

Liminal AI is a free, local-first coding agent — it edits your repo, runs tests and shell commands, uses git, and researches the web, all on your machine. Because it's model-agnostic, it has always supported any OpenAI-compatible provider. As of v0.0.16 it also supports running with no provider key of your own.

Option 1 — Bring your own key (Community Edition)#

This is the classic local-agent setup, and it stays completely free. You point Liminal at any OpenAI-compatible endpoint and pay that provider directly for tokens — Vireon takes no margin on your model spend.

• OpenRouter (recommended) — one key, hundreds of models, and prompt-cache discounts on supported routes.
• OpenAI / Anthropic — use your existing account directly.
• Local models — run fully offline through Ollama or LM Studio by pointing AGENT_API_BASE_URL at your local server.

Bring-your-own-key is the right choice when you already have a provider account, want a specific model, or need everything (including inference) to stay on infrastructure you control.

Option 2 — Managed inference (Pro)#

If you'd rather not manage a provider account at all, managed inference routes Liminal's model calls through Vireon's metered, OpenAI-compatible proxy. Sign in once and the agent just works — no key, no base URL, no separate billing relationship.

• No key setup — sign in with liminal login and the harness defaults to managed routing.
• Included credits — Pro comes with a monthly credit allowance at pass-through rates, with optional top-ups when you need more.
• Transparent usage — the web UI shows a live usage banner, and Account → Managed inference shows remaining credits.
• Switch anytime — flip back to bring-your-own-key in Settings whenever you want.

Even with managed inference, the agent still runs on your machine — only the model call is proxied. Your repository, shell, git state, and memory stay local.

Which should you choose?#

Both run the exact same harness — 140+ tools, persistent memory, sub-agents, and a live tool stream. The only difference is where the model call goes.

Get started in about 90 seconds#

1. Install (macOS / Linux / WSL)

   curl -fsSL https://www.vireondynamics.com/install/install.sh | bash
   

2. Install (Windows PowerShell)

   irm https://www.vireondynamics.com/install/install.ps1 | iex
   

3. Pick your path

   Add an OpenAI-compatible key in the wizard for Community, or run liminal login for Pro managed inference.

4. Open the UI

   Launch the web or terminal UI — same agent underneath.

FAQ#

Do I need an API key to use Liminal AI? Not necessarily. Community Edition is bring-your-own-key (paste any OpenAI-compatible key). On Pro, managed inference routes calls for you, so you can run Liminal with no provider key at all.

Is the agent still free? Yes — the Liminal software is free under FSL-1.1-MIT. With your own key you pay only your provider for tokens. Managed inference is a Pro feature with included monthly credits.

Does my code get sent to Vireon? No. The agent runs locally. With managed inference, only the model request is proxied; your repository, files, and session logs stay on your machine.

Can I switch between the two? Anytime, in Settings. Many people start on managed inference and later add their own key (or vice versa).