Codex

tokenolo is a drop-in local proxy for Codex. It's a middleware layer that sits between you and the agent — no rewiring, no new model, no change to how you work. You keep the same commands; tokenolo trims redundant filler from each message before it's sent, then gets out of the way.

How it fits

Codex assembles a prompt plus the surrounding context it would send to the model. tokenolo reads that input, removes redundant filler from your own messages (removal-only, so the text is always a subsequence of what you wrote), and forwards everything else — system prompt, tools, tool output, files, history — untouched.

prompt → Codex → tokenolo wrap → filler removed → model

The request goes to the model under your own key; nothing routes through a tokenolo server.

Setup

Note: the CLI is pre-release — confirm exact commands with tokenolo --help.

1. Install tokenolo on the machine where you run Codex:

   npm i -g tokenolo
   

2. Sign in. Create a free account at app.tokenolo.com/signup — no card required — generate an API key, then authenticate:

   tokenolo login --key tk_live_…
   

3. Wrap Codex. Launch it through tokenolo — this starts a local proxy on 127.0.0.1 and points Codex at it, so the filler in each message is trimmed on the way to the model:

   tokenolo wrap codex
   

4. Work as usual. Same commands you already use. tokenolo trims filler from each message on the way in, forwards the rest untouched, then steps aside.

What you get

• Private by design — your code never leaves your machine; requests go to the model under your own key, and nothing is stored.
• Cache-safe — byte-stable edits keep your prompt cache valid across turns.
• Zero rewiring — drop-in. No new model, no workflow change.

Next steps

• tokenolo wrap — the command reference.
• Claude Code — the same drop-in setup for Claude Code.
• What gets optimized — what tokenolo trims and what it forwards untouched.