Relaystation logo Relaystation

Recipes · bundle

Document Chores

OCR, convert, extract, transform — every mundane document and data job, one payment each

100+ compute chores behind one payTo: PDF (merge, split, OCR, render, watermark), Office↔PDF, images, audio/video, and a full CSV/data toolkit with SQL. Chain any sequence server-side in a single pipeline call — one payment, one result, no intermediate downloads. Reasoning is expensive; chores shouldn't be.

Cost per run: $0.01–0.06 typical (pipeline bills the sum of its steps)

Success rate: 5/5 (deterministic compute, single provider; pipeline is all-or-nothing per step with per-step receipts)

Services used

Service Provider Price Category
Pipeline (chain any ops) Relaystation $0/call + sum of steps Orchestration
PDF toolkit (merge/split/OCR/render/…) Relaystation $0.001–0.004 / page Documents
Office → PDF / markdown → docx/pptx Relaystation $0.0003/MB–$0.005 flat Office
Data & CSV (SQL, join, pivot, xlsx I/O) Relaystation $0.0002–0.0005 / MB Data
Images & media (convert/resize/trim/…) Relaystation $0.0003/MP–$0.0005/MB Media
Document AI (tables, forms, invoices, Q&A) Relaystation $0.005–0.08 / document Extraction

Workflow

The showcase in one call: POST /v1/bundles/ingest-document with a scanned document — OCRs it searchable, extracts the tables, writes them to xlsx, and drops both artifacts to durable URLs. Bills the sum of steps.

Manual alternative (step by step)

  1. Single chore: POST the matching /v1/{pdf,image,media,data,csv,doc,text}/... route directly
  2. Multi-step: POST /v1/pipeline with the step list — e.g. scanned PDF → ocr-searchable → extract-tables → data/to-xlsx → storage/drop — one call returns the final artifact URL
  3. Inspect before you process: probe/metadata/profile routes cost ~$0.0002 and prevent wasted steps
  4. Receipts itemize every step; sub-cent remainders auto-credit

Prompt

You are an agent that offloads mechanical work to Relaystation (x402, Base USDC; or an API
key) instead of doing it in-context. Rule of thumb: if the job is deterministic — convert,
resize, merge, OCR, reshape, query — it is a chore; buy it for a fraction of a cent rather
than reasoning through it. For "scanned document in, searchable PDF + spreadsheet out", use
POST /v1/bundles/ingest-document — one payment, both artifact URLs back. For any other
multi-step job, use POST /v1/pipeline so intermediate files never round-trip through you:
one payment, one final result URL. Before heavy processing, spend $0.0002 on the inspect
routes (media/probe, image/metadata, data/profile) to check the input is what you think it
is. For tabular work prefer /v1/data/sql (full SQL, sandboxed DuckDB) over chains of
filter/join/sort calls when the logic fits in one query. Failures return structured errors,
not charges for undelivered results. Each call is one x402 payment; sub-cent remainders
auto-credit to your balance.

← All recipes