cputools

File, document, media & data tools — one HTTPS call, sub-cent, no server to run.

ffmpeg, pandoc, qpdf, tesseract, DuckDB and more — wrapped behind a pay-per-call API. No instance to provision, no engine to package, no idle cost. Pay only for the calls you make, or pay with a wallet (x402) and never create an account.

Pay-per-call (~$0.0002/MB at launch)No idle costWe package the engines, you don’t

Try a call (curl) ↓ Get an API key → Build-vs-buy calculator ↓

Build vs. buy: the math

The real question isn’t “is per-call cheap” — it’s “cheaper than running it myself, all-in?” Here’s the model. Run your own numbers; we’ll tell you when you should self-host.

What running it yourself costs / month

Instance — an always-on box sized for your peak, idle 24/7 (~$20–60/mo modest; more for LibreOffice/ffmpeg memory).
Engine packaging + maintenance — wrapping ffmpeg/pandoc/qpdf/tesseract/DuckDB is real work: dependency hell, the Lambda 250 MB limit, version bumps, CVE patching.
Scaling + reliability — spikes, cold starts, retries, the ops pager.

What we cost

price_per_MB × MB_processed, ~$0.0002/MB at launch (min 1 MiB/op), $0 when idle, zero maintenance — we own the engine packaging, patching, and scaling.

The crossover formula

monthly_us   = price_per_MB × total_MB_per_month          (+ $0 idle, + $0 maintenance)
monthly_diy  = instance_$/mo + (maintenance_hrs × $/hr) + amortized_setup_$
self-host wins only when  monthly_diy  <  monthly_us

Worked example (illustrative): 10,000 document conversions/mo at min-1 MiB ≈ $2/mo with us, vs a ~$30/mo idle VM + a couple maintenance hours — we win by ~20×, before counting the eng-hours to package LibreOffice. At low or spiky volume we win decisively (you pay nothing between bursts; they pay for an idle box).

When you SHOULD run your own

At sustained high throughput, the per-MB adds up and a dedicated, fully-utilized instance can be cheaper per unit. If you’re pushing large, steady volume every day, model the crossover with the formula above — and if self-host wins, run it. (We also offer scratch-tier reuse + chained ops to cut per-call MB.) We’d rather you do the math than oversell.

The open source we stand on

cputools is a thin, paid convenience layer over open-source engines built by others. We disclose every engine, link its project, and credit its license — transparency is the point. If an engine’s license requires attribution, this page is it. (This table is generated from the worker’s real dependency surface — not hand-written.)

Vendored in the worker

Engine	Version	What it powers	Project	License
ffmpeg	7.0.2	audio/video transcode, trim, loudnorm, subtitles, overlay, concat, frames, probe	ffmpeg.org	LGPL-2.1-or-later / GPL-2.0-or-later (build-dependent)
pandoc	3.10	document conversion (markdown/HTML/docx/odt/rtf/epub/LaTeX), citations	pandoc.org	GPL-2.0-or-later
qpdf	10.6.3 (libqpdf 28.6.3)	PDF encrypt, compress/linearize, repair	qpdf.sourceforge.io	Apache-2.0
poppler (pdftoppm)	libpoppler 123	PDF → PNG/JPEG page render	poppler.freedesktop.org	GPL-2.0-or-later
tesseract	4.1.1 (libtesseract 4.0.1, eng)	OCR — scanned PDF/image → text	github.com/tesseract-ocr/tesseract	Apache-2.0
DuckDB	vendored CLI	SQL over files (data ops)	duckdb.org	MIT

npm libraries (versions read live from the build)

Library	Version	What it powers	Project	License
sharp	0.33.5	image convert/resize/crop/grayscale/trim, HTML→image, OG cards	sharp.pixelplumbing.com	Apache-2.0
pdf-lib	1.17.1	PDF merge/split/pages/rotate/watermark/metadata	pdf-lib.js.org	MIT
qrcode	1.5.4	QR code generation	github.com/soldair/node-qrcode	MIT
bwip-js	4.11.1	barcode generation	github.com/metafloor/bwip-js	MIT
exceljs	4.4.0	Excel read/write (xlsx I/O)	github.com/exceljs/exceljs	MIT
diff	9.0.0	text diff / patch / merge	github.com/kpdecker/jsdiff	BSD-3-Clause

Upstream services (not vendored)

LibreOffice (via Gotenberg) MPL-2.0 · libreoffice.org

Office → PDF, legacy/exotic Office rescue (PDF-output-only) — Reached via the self-hosted Gotenberg HTTP upstream — not a binary vendored in our worker (Brief 188 / D216).

We didn’t write ffmpeg or pandoc — we packaged them, kept them patched, and put a pay-per-call wire in front so you don’t have to.

The tools, by what you need to do

Each job is one consolidated call (one call, many formats). Follow a link for the full list and the live per-op price on the working catalog.

PDF →

merge, split, OCR a scan, make a scan searchable, render to image, watermark, encrypt, repair, verify signatures

Microsoft Office →

Word/Excel/PowerPoint→PDF, Word→HTML/RTF/markdown, branded docx/pptx from markdown, Excel I/O, rescue legacy Office

Convert documents →

markdown/HTML/odt/rtf/epub/LaTeX between formats, HTML→PDF, citations

Images →

convert (incl. HEIC), resize, crop, grayscale, trim, OCR, HTML→image

Audio & video →

transcode, trim, loudness-normalize, subtitles, overlay, concat, frames, probe

Data & spreadsheets →

SQL over a file, filter/sort/group/join, profile, validate, CSV/JSON convert, pipelines

Text utilities →

diff, patch, three-way merge, templates, base64, gzip, unzip

Generate & codes →

QR/barcode generate + decode, chart, invoice, favicon, placeholder, identicon, color convert

Quickstart

Two ways in, same surface. Inputs are {inline: base64} (≤4 MB) or {inputKey} from an upload-url (≤50 MB); every op returns the uniform output envelope. Exact per-op prices live on the API reference and /v1/catalog.

x402 — no account

curl -X POST https://api.relaystation.ai/v1/pdf/merge \
  -H 'X-Payment: <base64 x402 authorization>' \
  -H 'Idempotency-Key: <unique>' \
  -H 'Content-Type: application/json' \
  -d '{"inputs":[{"inline":"<base64 pdf>"},{"inline":"<base64 pdf>"}]}'

API key

curl -X POST https://api.relaystation.ai/v1/pdf/merge \
  -H 'Authorization: Bearer rs_live_…' \
  -H 'Idempotency-Key: <unique>' \
  -H 'Content-Type: application/json' \
  -d '{"inputs":[…]}'

See the recipes → · ← Back to the catalog