tokenolo wrap
Reference for the core wrap command.
tokenolo wrap is the command that does the work. It launches your coding agent
behind a local proxy on 127.0.0.1, then removes redundant filler from each of
your messages on the way to the model — removal-only, with everything else
(system prompt, tools, tool output, files, history) forwarded untouched.
Note: the options below map the current command surface. Run
tokenolo wrap --helpon your installed version for the authoritative flags and defaults.
Usage
tokenolo wrap claude
This starts a local proxy on 127.0.0.1 and launches Claude Code pointed at it.
Work exactly as you always do — type prompts, run tools, edit files. tokenolo
trims each message inline, then gets out of the way. No rewiring: it's a drop-in
for Claude Code.
The rewrite happens on 127.0.0.1, and it fails open — if anything is
unexpected, your original message passes straight through, untouched.
Options
<target>— the agent to wrap and launch:claudeorcodex. Required. tokenolo starts it behind the local proxy so every message is optimized in flight.[args…]— anything after the target is passed straight through to the agent (for exampletokenolo wrap claude --model opus).--dry-run— log what would be optimized but forward every prompt unchanged, so you can preview the rewrite without changing what the agent sends.--log <path>— also append the rewrite audit log to a file, so you can review what was trimmed on each message.--help— show the authoritative options and defaults for your installed version. This is the source of truth if anything here differs.
Examples
Wrap Claude Code for an optimized session:
tokenolo wrap claude
Preview what would be trimmed without changing anything the agent sends:
tokenolo wrap claude --dry-run
Wrap Claude Code and save the rewrite audit log to a file:
tokenolo wrap claude --log ./tokenolo.log
Wrap Codex instead:
tokenolo wrap codex
Related
- CLI overview — the full command surface at a glance.
- What gets optimized — what tokenolo removes and what it forwards untouched.