01What Reduce Tokens AI Agents Means

Reduce tokens AI agents means cutting the number of tokens your AI coding agents use. Tokens are the pieces of text the AI reads and writes. Fewer tokens means lower costs and faster performance.

This episode shows you free tools and methods to shrink token usage. You can use them with any AI system, not just Agent OS. The host, Julian Goldie, answers a community member's question about setting up Agent OS without wasting tokens.

02Free Tools to Cut Token Usage

Julian recommends four free, open-source GitHub repos. They all minimize tokens inside an agentic operating system. You can plug them into Codex or any other system.

  • Headroom: Reduces tokens by about 20% for coding agents.
  • Ponytail: Acts like a lazy senior developer, creating minimal code to save tokens.
  • Caveman: Cuts output tokens by about 65% by making replies less verbose.
  • RTK: Another repo that helps minimize token usage (exact details not covered).

These tools are free and work with any AI setup. You don't have to use Agent OS to benefit.

03How to Set Up Your Agent OS

Getting started with Agent OS is simple. Julian answers common setup questions. Follow these steps:

  1. Choose your coding tool. Julian currently prefers ChatGPT over Claude for Agent OS, but that can change month to month.
  2. Skip VPS. Julian doesn't recommend VPS because it complicates things and has security downsides.
  3. Install Agent OS on one device first. Make sure it works before adding more devices.
  4. Add Tailscale later to connect multiple devices. It takes about 10 minutes to set up.

There is a full training on using Agent OS with Tailscale in the classroom's daily updates section.

04Using Free APIs and Local Models

You can run Agent OS for free. Julian explains several options for free APIs and local models.

  • Hermes: Use free APIs inside Hermes. Julian has a separate profile for local models.
  • LFM 2.5 2.6b: A free API trained on Hermes agentically. It runs locally and was very fast on a Mac Studio.
  • DGX Spark: Use the local setup inside the local machine section, with free local models.
  • Gemma 4: A free local model that works inside the system, but you need a decent setup like DGX Spark or an RTX.
  • Free AI coder: Switch between free APIs like Nine Router, Omni Route, and a new free API on OpenRouter. You can also use OpenCode with DeepSeek v4 Flash.

These free options are covered in the training for a free Agent OS.

05Results and Comparisons from the Episode

Julian shares specific numbers on token savings. The tools have real, measurable effects.

  • Headroom reduces tokens by about 20% for coding agents.
  • Caveman cuts output tokens by about 65% by making replies less verbose.
  • Ponytail minimizes tokens by creating minimal code, similar to a lazy senior dev.

Julian also compared ChatGPT and Claude. A month ago, he would have said Claude. Now he says ChatGPT is better for Agent OS.

He notes that these changes happen quickly, like when Astra 6 was released last week.

He tested LFM 2.5 2.6b on a Mac Studio and found it incredibly fast, faster than expected.

06Tips, Limits, and Mistakes to Avoid

There are common mistakes you should avoid. Julian gives clear advice.

  • Don't use a VPS for Agent OS. It complicates things and is not great for security.
  • Don't jump straight to Tailscale. Set up Agent OS on one device first.
  • Don't assume the token minimization playbooks only work with Agent OS. They work with any system.
  • Don't lock into one tool. Claude vs ChatGPT changes quickly, so stay flexible.

For free local models, you need a decent setup like DGX Spark or an RTX. Otherwise, it won't work well.

Also, note that tools like Caveman make replies less verbose. That means you get shorter answers, but that's the point.

07Who Should Reduce Tokens AI Agents For

This is for anyone using AI coding agents, whether through Agent OS, Codex, or other systems. If you pay for tokens, cutting waste saves money.

It matters because token usage can get expensive. By using free tools and free APIs, you can run agents for no cost. This is especially useful for beginners or small businesses.

Julian's community member, Bhavya, wanted to avoid perfection and just start. These tips help you get going without overspending.

08Where to Get More Help and Training

Julian offers a full training program at airport for boarding.com. Inside the classroom, you get access to all his best trainings on agent OS and token minimization.

There's a full Agent OS system you can grab as a zip file. It includes a video tutorial and is updated regularly. There's also a new beginner-to-expert course that breaks down systems to get leads and customers.

You can join weekly coaching calls and ask questions. The community map helps you meet people in your local area building with AI agents.

For more, visit airport for boarding.com or check the link in the description.