AI Token Saver

AI Token Saver

Choose a suitable model.
Spend less effort guessing.

Get a model and effort recommendation for your task from the live Model Value Map, on the AI platforms you use. With ChatGPT alone it stays within ChatGPT. With more than one platform it suggests another one only when the savings are large, and always offers to stay where you are.

Guidance updates when you use the plugin. No local server or browser automation needed. Optional setup adds the guidance to your instructions and one refresh task per platform, at 1:00 PM local time on the first Tuesday of each month.

Install by pasting a prompt

Copy one prompt and paste it into the AI app you want it installed in. An app with a terminal, such as Claude Code or Codex, installs it and checks it. A chat app without a terminal answers with the exact clicks.

Install the AI Token Saver plugin for me. Source: https://github.com/robbagley-afk/ai-token-saver

1. Work out which app you are. Check whether it is already installed and say so instead of installing twice. If the retired ChatGPT and Codex edition (`ai-token-saver@ai-token-saver`, shown as AI Token Saver (retired edition) or plain AI Token Saver with only OpenAI guidance) is installed, say so. After the new plugin is installed and verified, show me how you will remove the retired edition with this app's own plugin uninstall command or menu, and remove it only after I say yes.
2. Show me the commands you will run, then run them:
   - Claude Code: `claude plugin marketplace add robbagley-afk/ai-token-saver` then `claude plugin install ai-token-saver-all@ai-token-saver`
   - Codex: `codex plugin marketplace add robbagley-afk/ai-token-saver` then `codex plugin add ai-token-saver-all@ai-token-saver` These commands write to ~/.codex. If the sandbox blocks them with “Operation not permitted”, request approval to run them outside the sandbox. If approval is unavailable, give me both commands to run in my own Terminal and wait until I say done before verification. After an attempted install, inspect the installed state before retrying to avoid duplicates.
   - Any other agent with a shell (Gemini CLI, Antigravity and similar): add `https://ai-token-saver.vercel.app/mcp` as a Streamable HTTP MCP server named ai-token-saver with no authentication, using the command or config file your own documentation gives. Then save https://raw.githubusercontent.com/robbagley-afk/ai-token-saver/main/plugins/ai-token-saver-all/skills/model-guidance-all/SKILL.md into your skills folder as model-guidance-all/SKILL.md. If you have no skills folder, tell me and stop after the server is added.
   - A chat app with no shell (ChatGPT, Claude, Gemini or Grok in a browser or phone): you cannot install it yourself. Give me the exact clicks to add a custom MCP server or connector with the address https://ai-token-saver.vercel.app/mcp and no authentication, and tell me if my plan or workspace admin may block it.
3. Verify. List the installed plugins or MCP servers and confirm AI Token Saver is there. If the get_guidance tool is available in this chat, call it once with my platform and show me the Revision. If it only loads in a new chat, say so.
4. Finish with one line: INSTALLED, ALREADY INSTALLED, CLICKS NEEDED, or FAILED with the exact error. Then tell me to start a new chat and ask "Recommend a model and effort for this task. Cite the Revision."

Rules: do not edit my instruction files or account settings, do not create scheduled tasks, do not install anything else, and if a command fails show the error and stop, except for the explicit sandbox permission recovery above. In that case request outside-sandbox approval or provide the manual commands, wait for completion, and verify before continuing.

Using the retired ChatGPT and Codex edition? The same prompt installs AI Token Saver and offers to remove the old one.

Use in ChatGPT

  1. Open Plugins. If AI Token Saver has been shared with you, install it there.
  2. For direct access, choose Add custom MCP server, enter AI Token Saver, paste the server URL below, choose No authentication, and create the plugin. Your workspace may require admin approval. This adds the live tools; the full package also supplies the guidance workflow.
  3. In a new supported chat, select AI Token Saver and ask: Recommend a model and effort for this task. Cite the current Revision.
Use in Codex
Download the plugin package and add it through your client's supported plugin interface. Repo marketplace instructions are in the README. Client and workspace support can vary.
Use more than one platform?
Tell AI Token Saver which platforms you use. At the start of a project it recommends another platform only when that one costs about half as much or less for the same quality. For a long project it can suggest doing each phase on the platform that fits it, with the number of handoffs. Once you have started, it suggests a switch only for large savings, and it always gives you the option to stay. In Claude Code run /plugin marketplace add robbagley-afk/ai-token-saver and then /plugin install ai-token-saver-all@ai-token-saver. In Codex add the same repository as a marketplace and install AI Token Saver. In any other client that supports MCP, add the server URL above with no authentication and ask for guidance by platform name.
Optional setup
Ask the plugin to set up your instructions. It asks which platforms you use and gives you one prompt, with the app to paste it into. That app sets up every platform it can reach on your computer. For the rest, each app writes the handoff prompt for the next. Each platform gets one instruction block and one task, AI Token Saver guidance refresh, that runs at 1:00 PM local time on the first Tuesday of each month and refreshes the block when the guide changes. Existing guidance-refresh tasks are found first and only one stays active. Tasks that publish a guide are never touched. Nothing is edited without your yes.
What this does
Recommends settings and cites the source Revision. You select the model yourself. Savings depend on the task. Benchmark API prices are not subscription usage estimates. The plugin does not automatically run on every prompt.
Explore the source guide · Source code