Stops the re-read
The agent stops pulling in the same file it already has. Re-read suppression catches the repeat before it costs you tokens.
The one-time tune-up for your AI coding tools
It runs on your machine, trims what the agent wastes, and writes a receipt of the net. Buy it once.
curl -fsSL https://usetare.dev/install.sh | sh
tare demo shows your first receipt in about 60 seconds.
01 Sessions go dumb halfway through, and you can't say why.
02 It read the same file three times in one task.
03 Limits hit with no warning, mid-thought.
04 Bills nobody can explain, stacked on top of each other.
How it works
The agent stops pulling in the same file it already has. Re-read suppression catches the repeat before it costs you tokens.
Oversized tool output stops clogging the context window. The full output spills to disk, recoverable whenever you need it.
As the context window fills, Tare adapts — trimming harder when it's tight, staying light when there's room.
An append-only on-disk ledger records the tokens saved — measured numbers kept separate from estimated ones, so the proof is yours to check.
It also compresses around 100 shell commands, shows your savings and rate-limit pacing right in the Claude Code statusline, and ships with a local self-tuning Lab that improves the trim policy against your real work — all on your machine.
Proof
Every saving lands in an append-only local ledger. Measured numbers stay
separate from estimated ones — never mixed. You can read it back yourself
with tare report and verify it with tare evidence.
A 30-day measured number is coming. Until then, here's one real run.
Scope
We won't claim a tool until we've certified it. Roadmap stays roadmap.
Pricing
Pay once. Own it forever. No subscription.
All sales final; no money-back guarantee (refunds only where the payment provider or law requires).
FAQ
Quiet local hooks for AI coding tools. Tare stops the agent re-reading the same file, trims bloated tool output that clogs the context window, eases up as context fills, and writes an append-only on-disk ledger proving the tokens saved. To tare a scale is to zero out the weight of the packaging — Tare zeroes out the wasted tokens and writes you a receipt of the net.
No. Tare runs entirely on your machine — zero egress by construction. The ledger, the spill files, the Lab: all local. Nothing is uploaded anywhere.
Honest answer: the value scales with usage. The more the agent reads, repeats, and floods context, the more there is to trim. If you use it lightly, there's less to save. You don't have to take our word for it — your own ledger shows your real number, run by run. There's no money-back guarantee, so read your receipts and decide for yourself.
Claude Code today, certified. Other harnesses are on the roadmap and will be added one at a time as we certify them — included in your one-time purchase. We won't claim a tool before it's ready.
Signed, lifetime, and free. Buy once and every update — including every harness we certify later — comes with it. No upgrade tier, no renewal.
The full tool: re-read suppression, output trimming with recoverable spill, pressure-adaptive easing, the append-only ledger, ~100 compressed shell commands, the Claude Code statusline, and the local self-tuning Lab. Lifetime updates. Local-only. $99 for the first 100 buyers, then $149.
All sales final; no money-back guarantee (refunds only where the payment provider or law requires). Run tare demo and read your own ledger before you buy — the proof is on your machine.