A free AI plugin for Unreal Engine: what it costs to run

How a free, open-source Unreal AI plugin differs from subscription and credit plugins, what you pay your model provider, and the three ways to keep that cost low or at zero.

6 min read

Two ways an AI plugin is sold

Most AI plugins for Unreal are sold as a service. You pay a monthly plan, a credit balance, or both, and your requests go through the vendor, which resells model access and adds its own margin and limits.

Kismet is the other kind. The plugin is free and open source, including for commercial work, and it does not sell model access at all. You connect it to a model provider you choose, and requests go from your editor straight to that provider. Kismet takes no payments and no account is needed for local use.

What changes for you

  • Trying it: a subscription AI plugin typically needs an account, and often a card, before the agent runs. Kismet is a download.
  • Running out: a credit plan can stop mid-task until you buy more. With your own provider there is no Kismet meter.
  • Models: a subscription AI plugin typically gives you the vendor’s hosted models. With Kismet you pick the provider and model, and can change them per task.
  • Your project data: requests go to the provider you chose. No Kismet server is in that path.
  • The code: the plugin source is public, so you can read what it does, build it yourself and change it.

The trade is that setup is yours. You create a key with a provider, or run a model locally, and you carry the provider’s bill directly.

What you pay a provider

Hosted providers bill by tokens: the text sent to the model and the text it writes back. An agent sends the conversation, the tool descriptions and the tool results on every step, so cost follows how many steps a task takes and how much it reads, not how many messages you type.

  • Model choice matters most. Within each provider’s section of the model menu, the strongest model is listed first and the fastest, cheapest one last.
  • Effort and thinking settings raise or lower how much the model writes before it acts.
  • Short, focused conversations cost less than one long conversation that covers many unrelated tasks.
  • The Kismet panel shows usage cost estimates computed from the token counts your provider reports. Your provider’s bill is the authoritative figure.

Kismet works with Anthropic, OpenAI, Google, DeepSeek, Z.ai, Moonshot and MiniMax with your own key, plus OpenRouter, local runtimes and custom endpoints. Current prices are on each provider’s own pricing page.

Three ways to pay less, or nothing

  1. 1.

    Use an API key with a cheaper model

    Keep a strong model for hard, multi-asset work and switch to a fast one for small edits and questions. A key belongs to a provider, not a model, so one key covers every model in that provider’s section.

  2. 2.

    Use a subscription you already have

    Kismet runs a local MCP server. Run Claude Code or Codex yourself with your existing plan, connect it to Kismet, and it works in the editor through Kismet’s tools. Kismet never signs in to those accounts.

  3. 3.

    Run a local model

    Ollama and LM Studio need no key and nothing leaves your machine. Results depend heavily on model size, so read the local model guide before relying on this for real work.

Getting started

  1. 1.

    Install the plugin

    Quit Unreal Editor, then run the installer for your system. It installs Kismet for every supported engine it finds.

    # macOS
    curl -fsSL https://kismetai.io/install.sh | bash
    
    # Windows (PowerShell)
    irm https://kismetai.io/install.ps1 | iex
  2. 2.

    Add a key

    Open Kismet from the editor toolbar, open the model menu and click Add key on a provider’s section. Keys are stored encrypted on your machine and are never sent to Kismet.

  3. 3.

    Ask for one small change

    Pick something you can check by eye, let the agent compile, and review the result before asking for more.

In the docs

Try it in your own project.

Kismet is free and open source. Install the plugin, connect a model, and ask for one small change.

Related guides