Migrate from Reducto¶

If you call Reducto today, you can move to ReadAnyDoc without rewriting your integration. ReadAnyDoc's server speaks Reducto's v3 HTTP API, and Reducto's own SDKs work against it with one changed URL.

Why teams switch¶

	Reducto	ReadAnyDoc
Price per page	$0.010 (r-1)	about $0.003 to $0.005 (engine), $0.001 to $0.002 (model only), paid to your model provider
Where documents go	Reducto's cloud	your machine, and the model provider you choose under your account
Model	Reducto's own r-1	Claude Haiku 5.5 by default; any Claude, OpenAI or self-hosted model
Source	closed	open source, Apache-2.0
Accuracy	see Benchmarks: ahead on layout, formulas, header removal and speed	ahead on text and multi-column pages, level on tables and forms

Step 1: run ReadAnyDoc¶

pip install "readanydoc[engine]"
export ANTHROPIC_API_KEY=sk-ant-...
export READANYDOC_API_KEY=choose-a-long-random-key     # so only your services can call it
readanydoc serve --host 0.0.0.0 --port 8000

Or use Docker.

Step 2: point your client at it¶

Python SDKEnvironment variablePlain HTTP

from reducto import Reducto

client = Reducto(
    api_key=os.environ["READANYDOC_API_KEY"],
    base_url="https://readanydoc.internal.example.com",   # was Reducto's
)
# everything else stays the same
result = client.parse.run(input=upload.file_id)

# the Python SDK reads this when base_url is not given
export REDUCTO_BASE_URL=https://readanydoc.internal.example.com
export REDUCTO_API_KEY=$READANYDOC_API_KEY

Replace Reducto's host with yours; keep paths, bodies and the Authorization: Bearer header.

Step 3: map the model setting¶

`settings.model` you send	What ReadAnyDoc runs
`r-1` (or nothing)	the configured profile, `engine` by default
`r-1-mini`, `legacy`	its cheaper tier: `engine` becomes `model-only` at medium effort

usage_breakdown.parse_model reports the model you asked for, as Reducto does; usage.model_usage says what actually ran and the tokens it used.

Step 4: check the differences that matter to you¶

Read Differences and limits, then run a sample of your own documents through both and compare. Most teams check:

the fields they extract, against their own answer key;
documents with forms, handwriting or complex layout, where Reducto is still stronger on layout;
latency: ReadAnyDoc takes 20 to 50 seconds a page with the engine against about 5 on Reducto.

What stays the same¶

Request and response shapes, reducto:// file ids, jobid:// chaining, sync and async routes, GET /job statuses, webhooks, error bodies, citations in Reducto's shape, chunking rules. Route-by-route detail: Compatibility.