Skip to content

TypeScript

@anyrow/sdk: every endpoint, typed, for Node.js, Bun, Deno, Cloudflare Workers and browsers.

@anyrow/sdk runs anywhere fetch does. Its types come from the API's OpenAPI document, so every parameter, body and response is checked as you write the call.

Install

npm install @anyrow/sdk

Create a client

import { AnyrowSDK } from "@anyrow/sdk"const anyrow = new AnyrowSDK({  baseURL: "https://api.anyrow.ai",  headers: { Authorization: `ApiKey ${process.env.ANYROW_API_KEY}` },  timeout: 30_000, // milliseconds; no timeout when unset})

Create the client in server code: the key acts as its project. In Cloudflare Workers, create it inside the request handler from env.

Make a call

An operation's id is its method. Path parameters go in params, query parameters in search, JSON bodies in json and form bodies (files) in form:

const { data, error, status } = await anyrow.row.list({  params: { project_id: "proj_...", table_id: "tbl_..." },  search: { limit: 50 },})

Errors

A call returns { data, error, status, response }: data on success, error (the error body) otherwise. Check error before using data:

const { data: table, error } = await anyrow.table.get({  params: { project_id: "proj_...", table_id: "tbl_..." },})if (error) {  console.error(error.error_key, error.message)} else {  console.log(table.name)}

To throw instead, set throwOnError: true on the client: a call then returns the data, and a failure throws a ClientError subclass (UnauthorizedError, ForbiddenError, NotFoundError, RateLimitError and one per status) with status and the parsed body in data.

import { AnyrowSDK, ForbiddenError } from "@anyrow/sdk"const anyrow = new AnyrowSDK({  baseURL: "https://api.anyrow.ai",  headers: { Authorization: `ApiKey ${process.env.ANYROW_API_KEY}` },  throwOnError: true,})try {  const table = await anyrow.table.get({ params: { project_id: "proj_...", table_id: "tbl_..." } })} catch (error) {  if (error instanceof ForbiddenError) console.error("The key lacks tables:read")  throw error}

Read every row

const rows = []let cursor: string | undefineddo {  const { data, error } = await anyrow.row.list({    params: { project_id: "proj_...", table_id: "tbl_..." },    search: { limit: 100, cursor },  })  if (error) throw error  rows.push(...data.rows)  cursor = data.nextCursor ?? undefined} while (cursor)

Upload a file

Put a Blob in form and the SDK sends multipart form data:

import { readFile } from "node:fs/promises"const { data: batch } = await anyrow.extract.table.once({  params: { project_id: "proj_...", table_id: "tbl_..." },  form: { file: new Blob([await readFile("./invoice.pdf")], { type: "application/pdf" }) },})

In a browser, pass the File from an <input type="file"> as it is.

Stream an extraction

The stream calls return an async iterable of server-sent events:

for await (const event of anyrow.extract.table.stream({  params: { project_id: "proj_...", table_id: "tbl_..." },  form: { text: "Invoice 42 from Acme, total 120 EUR, due 2026-11-01" },})) {  console.log(event.event, event.data)}

Verify webhooks

Deliveries follow Standard Webhooks; verify them with the standardwebhooks package as the Webhooks guide shows.