tokenolo
Get started

Trim your Claude Code prompts locally — your code never sent to us

tokenolo runs a local proxy that strips redundant filler from your prompts before Claude Code sends them - removal-only, cache-safe, your own key. Tool output, files and history pass through untouched.

npm install -g tokenolo
get started freeread the docs
Built for engineers who work with the tools you know

Invisible by design.

tokenolo is a middleware layer — not a new model and not a new workflow. It wraps your agent behind a local proxy, trims the filler from each message on its way out, then steps aside.

01

Run one command

tokenolo wrap claudestarts a local proxy on 127.0.0.1 and launches Claude Code pointed at it.

02

Work like you always do

Type prompts, run tools, edit files — nothing about your workflow changes.

03

tokenolo trims the filler inline

On the way out, each message is stripped of redundant filler — removal-only, on your machine, under your own key. Files, tool output and history pass through untouched.

Also works with Codex: tokenolo wrap codex

Before and after.

The filler trim is live today. Context and tool-output pruning are on the roadmap — shown here with example figures, not measured results.

LiveFiller trimmingThrowaway politeness stripped from your prompt — removal-only, meaning preserved.
Raw input
please, could you update the readme
when you get a chance — thanks!
Via tokenolo
update the readme

Removal-only · meaning preserved · the “after” is always a subsequence

RoadmapVerbose tool outputA 1,400-token directory dump the agent only needs three lines of.
Raw input
$ ls -la node_modules/.bin
total 4128
-rwxr-xr-x  1 u  staff  384  acorn
-rwxr-xr-x  1 u  staff  384  acorn-walk
… 212 more entries …
Via tokenolo
# 215 bin entries — relevant to this task:
next   eslint   tsc
# full listing pruned

Same next step chosen · 1,402 → 38 tokens · example

RoadmapUnread contextThe agent attaches every open file; the task reads a fraction of them.
Raw input
# context assembled by the agent
system prompt + open files + history
42 files attached · ≈ 1,842 tokens
Via tokenolo
# pruned to what this task reads
9 relevant files kept · 33 dropped
≈ 612 tokens

Same answer · 1,842 → 612 tokens · example

Simple pricing. Scale on wraps.

Every plan has the same features - they differ only by your daily wrap limit and rate. Pick a plan, raise the limit when you grow.

Free
$0/mo
No card required
  • 3 wraps / day
  • 20 requests / min
  • Every feature, no gates
Start free
Pro
$19/mo
Billed monthly
  • 50 wraps / day
  • 60 requests / min
  • Every feature, no gates
Get started
Team
$49/mo
Billed monthly
  • 500 wraps / day per seat
  • 120 requests / min
  • Every feature, no gates
Get started
Business
$99/mo
Billed monthly
  • Unlimited wraps per seat
  • 300 requests / min
  • Every feature, no gates
Get started
Enterprise

Custom wrap + rate limits, SSO / SAML & audit logs, and dedicated support with an SLA. Built for teams running tokenolo at scale.

Talk to sales

Questions? Answered.

What is tokenolo?
A local proxy for Claude Code. It runs on your machine (127.0.0.1) and forwards every request to Anthropic with your own key. The only thing it changes is your prompt text, where it removes redundant filler - removal-only, so meaning is preserved. Your files, tool output and history are forwarded untouched, and nothing is stored.
Which coding agents does it work with?
Drop-in for Claude Code. tokenolo is a middleware layer, not a new model - it slots into the agent you already run.
Do I have to change my workflow?
No rewiring. Same agents, same commands. tokenolo optimizes the input before it is sent, then gets out of the way.
How much does it save?
Modestly - it removes filler, not context. The real value is a private, cache-safe trim with zero workflow change: your code never leaves your machine, and your prompt cache stays valid across turns.
What's on the roadmap?
Deeper input optimization - pruning unread context and dead files, and trimming oversized tool output before send. That work is in development; today tokenolo ships the local, removal-only filler trim.
How much does it cost?
Pro $19/mo, Team $49/mo, and Business $99/mo - plans differ by your daily wrap limit and rate. See Pricing for the full breakdown.
How do I get started?
Sign up, point tokenolo at your task, and run your agent as usual. Want to see it first? Book a demo.

Your next run starts lean.

Drop tokenolo in, keep every command you already run, and watch the token bill fall - same model, comparable results.

Get started
HumansMachine