TypeScript / Node.js API

import {
  preprocess,
  preprocessString,
  preprocessWithOptions,
  preprocessGmail,
  toLlmContext,
  toLlmContextWithOptions,
  RenderMode,
} from "langmail"
import type {
  ProcessedEmail,
  PreprocessOptions,
  LlmContextOptions,
  Address,
  CallToAction,
  ThreadMessage,
} from "langmail"

preprocess()

Accepts raw RFC 5322 email bytes and returns a structured ProcessedEmail object. Handles MIME multipart messages, HTML and plain-text body variants, and normalises character encodings.

function preprocess(raw: Buffer): ProcessedEmail
function preprocessString(raw: string): ProcessedEmail
function preprocessWithOptions(
  raw: Buffer,
  options: PreprocessOptions
): ProcessedEmail

preprocess is synchronous and takes a Buffer. Use preprocessString as a convenience wrapper if you already have the email as a string. Use preprocessWithOptions to override defaults — see PreprocessOptions.

ProcessedEmail

Optional fields are declared with ?: in the generated .d.ts — their type is T | undefined, not T | null.

Address is { name?: string, email: string }. CallToAction is { url: string, text: string, confidence: number }. ThreadMessage is { sender: string, timestamp?: string, body: string }.

PreprocessOptions

preprocessGmail()

Provider adapter for the Gmail API. Accepts the response of gmail.users.messages.get({ id, format: "full" }) from googleapis and returns the same ProcessedEmail shape as preprocess(). Skips MIME re-parsing — the Gmail API has already decomposed the message into typed parts, so the adapter walks payload.parts, base64url-decodes the bodies, and feeds them into the shared cleaning pipeline. The body tree walk, header parsing, and base64url decoding all happen in Rust — this wrapper only serializes the caller's object to JSON and delegates to the native binding, so the output is byte-identical to the Python and Rust entry points.

function preprocessGmail(
  msg: GmailInput,
  options?: PreprocessOptions
): ProcessedEmail

Accepts either the bare Schema$Message or the full googleapis response ({ data: Schema$Message, ... }). The message must have been fetched with format: "full" so payload is present with headers and base64url-encoded body parts.

Body selection: walks payload.parts depth-first and picks the first non-attachment leaf of each type. When both text/html and text/plain are present, HTML wins. Parts with Content-Disposition: attachment or a filename are skipped.

Throws:

• TypeError if the input is not an object or has no payload (i.e. the message wasn't fetched with format: "full").
• Error if the chosen body part is attachment-backed (Gmail returned body.attachmentId instead of body.data because the body exceeded the inline size threshold — fetch with users.messages.attachments.get and inline the decoded content).

toLlmContext()

Accepts a ProcessedEmail and returns a deterministic plain-text string formatted for direct inclusion in a prompt. The output includes a header block (FROM / TO / SUBJECT / DATE) followed by a CONTENT: section.

function toLlmContext(email: ProcessedEmail): string
function toLlmContextWithOptions(
  email: ProcessedEmail,
  options: LlmContextOptions
): string

Use toLlmContextWithOptions when you need to control rendering — for example, to include quoted reply history.

LlmContextOptions