A local tool that shrinks documents for AI agents

Docpipe

Stop paying full price every time your AI reads a document.

AI models bill by the token — roughly a word fragment — and they bill every single time they read a file. Docpipe runs on your machine and shrinks large PDFs, Markdown, and HTML into compact packages your agent works from instead. In our measured tests that cut 54–87% of the tokens, while numbers, IDs, dates, and quotes are kept word-for-word and the original stays one step away.

install Frugal docpipe skill install --all --json

-- fewer tokens across our local document test set

102,856 tokens saved on one real research PDF

-- exact facts preserved word-for-word

See the measured results Estimate your savings

docpipe - one real run, one real PDF

$ docpipe pack ./2607.12056.pdf --profile prompt_light --out .docpipe-demo
reading: 189,855 tokens of source PDF
package: 86,999 tokens — 54.2% smaller
kept exact: 176 facts (numbers, IDs, quotes) word-for-word
recovery: every summary points back to the original text
result your agent reads the small package instead of the full PDF.

Measured results — the charts first

Real token reductions from real files.

Each pair of bars below is one of our own project documents: the top bar is what it costs to paste the raw file into an AI conversation, the bottom bar is the Docpipe package for the same file. Smaller is cheaper. We also show the one case where Docpipe loses — tiny files — because you should know when not to use it.

fewer tokens overall across the test set of five real document batches — including the small-file batch where Docpipe adds overhead instead of saving.

Why this matters: agents re-read documents constantly — every follow-up question, every handoff to another agent, every retry. A file that costs 10,000 tokens raw and 1,400 packaged saves you roughly 86% on every one of those reads.

Loading test details...

One real PDF, before and after

An 11-page research paper: 54.2% cheaper to read.

We ran a real research paper (2607.12056.pdf) through Docpipe. The agent gets a package that is roughly half the tokens of the raw PDF. The full original stays on disk for checking and recovery — it just never has to be loaded into the conversation again.

Raw PDF costs 189,855 tokens per read

Docpipe package costs 86,999 54.2% less, every read

Nothing important lost 176 facts kept word-for-word

Token economy chart comparing PDF source, prompt-light payload, dense PNG pages, and full archival package

102,856 tokens saved on every single read of this document.

5 / 11 pages were bulky enough to be carried as compact page images ("visual gist") instead of full text — a further 75.3% reduction for background context, never used for exact facts.

Every package gets a stable ID (pdf_ccd97b569aabb51f) so another agent — or you, next week — can pick it up without re-reading the source.

Contact sheet of five dense PNG pages from the prompt-light package — The 5 pages carried as compact images

Dense PNG sheet showing pages 0001 through 0003 — Sheet 01 / pages 1–3

Dense PNG sheet showing pages 0004 through 0005 — Sheet 02 / pages 4–5

Page images are for background context only, never for exact facts. Anything exact stays as text with a link back to the original.

Estimate your own savings

Slide to your file size. See what Docpipe saves.

Drag the slider to a document size and pick the file type. The estimate uses the same measured rates as the charts above — including the honest answer that tiny files are cheaper to read raw.

Document size 120 KB

File type

Times your agents read it 10×

Raw file --

Docpipe --

Tokens saved per read --

Reduction --

Saved across 10 reads --

≈ cost saved at $3 / million tokens --

Move the slider to see an estimate.

Estimates use the measured rates on this page: 84.1% for large text documents, 54.2% for PDFs, and the measured overhead on tiny files. One token ≈ 4 characters. Your documents and model prices will vary — the point is the shape, not the penny.

How it works

Install once. Point it at a file. Read the small version.

Docpipe is a single program that runs on your computer. Your documents never leave your machine — there are no network calls in the conversion path.

01

Install the binary and Frugal skill once

Download Docpipe, then install the agent skill named frugal. The CLI stays docpipe; Frugal is the agent-facing instruction surface for Claude Code, Codex, and MCP.
02

Docpipe splits the document into three channels

Exact facts — numbers, IDs, dates, quotes — stay as text, word-for-word, always. Bulky background gets compressed hard. Everything keeps a handle pointing back to the original, so nothing is ever truly gone.
03

Your agent reads the small package

Every read of the package instead of the raw file is tokens — and money — saved. When another agent picks up the work, it receives the package ID, not a 190,000-token paste.

When to use it

Use it when the document is bigger than the task.

< 8 KB	Don't bother — reading the raw file is cheaper. Docpipe's own tests show packaging tiny files adds ~17% overhead.
8–16 KB	Worth it when the file is document-shaped (reports, specs) or when exact values must survive.
> 16 KB	Almost always worth it — our large-document tests show 84–87% savings.

Also use it whenever exactness matters more than size: hashes, contract clauses, citations, and amounts survive packaging word-for-word — unlike an AI's summary from memory.

Use the frugal skill for this document.

The safety rule behind everything

Exact facts never ride in images.

Kept sharp

IDs, amounts, quotes, clauses

Anything where a single wrong character matters stays as real text your agent can quote and compare safely. This rule is enforced in code, not policy.

Compressed hard

Bulky background context

Long descriptive passages become compact summaries or page images — good enough to understand the gist, cheap enough to carry everywhere.

Always recoverable

The original, one step away

Every compressed piece points back to the exact original text. If a task suddenly needs the fine print, the agent fetches just that piece — not the whole file.

Output formats

Pick an outcome, not a config file.

You ask for what you want in plain language. Available today: clean Markdown conversion. The rest are on the public build roadmap below, gated behind measured verification — not shipped until proven.

Give me Markdown

Available now Turn a document into clean Markdown plus a source map that records where every piece came from.

Build a prompt

Coming next A ready-to-paste bundle: instructions, contents, exact facts, and key excerpts, sized for a model's context window.

Token-light package

In progress The smallest possible input — like the PDF demo above. Deliberately lossy, so it unlocks only for AI models that pass measured verification.

RAG assets

Planned Chunks, metadata, and source maps ready for search and retrieval systems. Anything inferred is labeled as inferred.

Audit package

Planned Full internals and diagnostics for QA, legal, and compliance review — see exactly what was kept, compressed, and why.

Install

One download, one command.

Docpipe ships as a normal application download: a platform binary with a checksum you can verify. No account, no cloud, no build-from-source.

Install once, then the skill routes documents by itself.

The binary installs the agent skill, so the instructions your agent follows and the tool it calls always match versions.

Runs entirely on your machine — no network calls while converting.
Same input always produces the identical package, byte for byte.
Exact values always cite the page and block they came from.
Lossy modes stay locked until the target AI model passes verification.

Download the binary for your platform

macOS, Windows, and Linux archives with a checksum manifest. (Code signing is a separate release step still in progress.)

download docpipe for macOS, Windows, or Linux

Teach Frugal to your agents

This emits the agent skill named frugal into Claude and Codex while the local command remains docpipe.

docpipe skill install --all --json

Run any document through it

docpipe pack ./2607.12056.pdf --profile prompt_light --out .docpipe-demo

Your agent gets the compact package; exact facts and the recovery path come along for free.

Live build status

This page reads its numbers from the actual build.

Every figure above comes from generated data files, not hand-written marketing copy. Here is where the project actually stands.

Current phase

Status pending

The static fallback stays quiet until generated status data is present.

Working today

profilesList supported output and model profiles.

Built with

docpipe-cli
rust toolchain pending

Token test baseline

Baseline data loads from eval/ab/baseline.json.

Build phase	Gate	Status
local	pending	Generate status.json for the live table.

Get started

Download once. Save tokens on every read after.

Install Docpipe, teach the frugal skill to your agents with one command, and let every large document travel as a compact package with its exact facts intact.