Jina Reader

Jina Reader fetches a page and returns it as clean markdown or HTML. The API is header-driven: the URL is the path and every option is a request header, so it is the lightest backend to run. Pricing is pay-as-you-go by token, which suits high-volume reading.

The @effect-uai/jina package already ships Jina embeddings; the reader is a second tag on the same package.

Install

pnpm add @effect-uai/core @effect-uai/jina effect

Wire it up

import { Config, Effect, Layer } from "effect"
import { FetchHttpClient } from "effect/unstable/http"
import { layer as jinaLayer } from "@effect-uai/jina/JinaReader"

const provider = Layer.unwrap(
  Effect.gen(function* () {
    const apiKey = yield* Config.redacted("JINA_API_KEY")
    return jinaLayer({ apiKey })
  }),
)

const mainLayer = provider.pipe(Layer.provide(FetchHttpClient.layer))

jinaLayer registers two service tags from one implementation:

• JinaReader: the typed tag. Yield this for the Reader headers below.
• WebRead: the generic tag for provider-portable code.

Calling it

import { JinaReader } from "@effect-uai/jina/JinaReader"

const program = Effect.gen(function* () {
  const reader = yield* JinaReader
  return reader.read({
    url: "https://effect.website/docs",
    format: "markdown",
  })
})

Every field from CommonReadRequest works here. The typed request adds:

Field	Type	Meaning
engine	browser curl auto	Fetching engine. browser renders JS.
targetSelector	string	CSS selector to extract instead of the page.
noCache	boolean	Bypass Jina’s page cache.
withLinksSummary	boolean	Return the page’s links on links.

Links are opt-in (withLinksSummary) because they add tokens to the response, which is what Jina bills on.

See also

• Web reading: the cross-provider concept and the portable request / response shapes.
• Market intel: read plus structured extraction, runnable against Jina with READ_PROVIDER=jina.