feat(account): add managed credits and cloud gateway

Introduce optional Apple account-backed credits with scoped gateway access while preserving local and BYOK paths. Refresh assistant behavior, tests, privacy disclosures, docs, and the website for the 2.0 experience.
This commit is contained in:
Rocky
2026-08-20 11:43:21 +08:00
parent 0f9280bd00
commit 498f407585
301 changed files with 19221 additions and 10891 deletions
@@ -0,0 +1,256 @@
// GatewayGrantCoordinator.swift
// OSGKeyboard · Shared
//
// Creates and rotates scope-limited gateway grants. Refresh rotation is merged
// inside this actor so concurrent extension requests never replay an old token.
import CryptoKit
import Foundation
public actor GatewayGrantCoordinator {
public static let defaultBaseURL = URL(string: "https://account.osglab.com")!
private let baseURL: URL
private let store: any GatewayGrantCredentialStore
private let session: URLSession
private let now: @Sendable () -> Date
private var refreshTask: Task<ManagedGatewayGrantCredentials, Error>?
public init(
baseURL: URL = GatewayGrantCoordinator.defaultBaseURL,
store: any GatewayGrantCredentialStore = GatewayGrantKeychainStore(),
session: URLSession = .shared,
now: @escaping @Sendable () -> Date = Date.init
) {
self.baseURL = baseURL
self.store = store
self.session = session
self.now = now
}
/// Host-only integration point. The account access token authorizes grant
/// creation but is used only for this request and is never persisted here.
@discardableResult
public func createGrant(
accountAccessToken: String,
scopes: Set<ManagedGatewayCapability>,
lifetimeSeconds: Int? = nil,
idempotencyKey: String = UUID().uuidString
) async throws -> ManagedGatewayGrantCredentials {
guard !accountAccessToken.isEmpty else { throw ManagedGatewayError.invalidGrant }
guard !scopes.isEmpty else {
throw ManagedGatewayError.server(
code: "invalid_request",
status: 400,
requestId: nil
)
}
struct Body: Encodable {
let scopes: [ManagedGatewayCapability]
let lifetimeSeconds: Int?
}
var request = URLRequest(url: endpoint("v1/gateway/grants"))
request.httpMethod = "POST"
request.setValue("application/json", forHTTPHeaderField: "Content-Type")
request.setValue("Bearer \(accountAccessToken)", forHTTPHeaderField: "Authorization")
request.setValue(idempotencyKey, forHTTPHeaderField: "Idempotency-Key")
request.setValue(UUID().uuidString, forHTTPHeaderField: "X-Request-ID")
request.httpBody = try JSONEncoder().encode(
Body(scopes: scopes.sorted { $0.rawValue < $1.rawValue }, lifetimeSeconds: lifetimeSeconds)
)
let credentials = try await sendGrantRequest(request)
guard credentials.scopes == scopes else {
throw ManagedGatewayError.invalidGrant
}
try await store.save(credentials)
return credentials
}
public func accessToken(
for scope: ManagedGatewayCapability,
forceRefresh: Bool = false
) async throws -> String {
guard let credentials = try await store.load() else {
throw ManagedGatewayError.missingGrant
}
guard credentials.scopes.contains(scope) else {
throw ManagedGatewayError.scopeNotGranted(scope)
}
if !forceRefresh, credentials.hasUsableAccessToken(for: scope, at: now()) {
return credentials.accessToken
}
guard credentials.hasUsableRefreshToken(at: now()) else {
try? await store.delete()
throw ManagedGatewayError.invalidGrant
}
return try await refresh(credentials).accessToken
}
public func clearGrant() async throws {
refreshTask?.cancel()
refreshTask = nil
try await store.delete()
}
private func refresh(
_ credentials: ManagedGatewayGrantCredentials
) async throws -> ManagedGatewayGrantCredentials {
if let refreshTask {
return try await refreshTask.value
}
let task = Task {
try await requestRefresh(using: credentials)
}
refreshTask = task
do {
let refreshed = try await task.value
refreshTask = nil
return refreshed
} catch {
refreshTask = nil
throw error
}
}
private func requestRefresh(
using credentials: ManagedGatewayGrantCredentials
) async throws -> ManagedGatewayGrantCredentials {
struct Body: Encodable {
let refreshToken: String
}
var request = URLRequest(url: endpoint("v1/gateway/grants/refresh"))
request.httpMethod = "POST"
request.setValue("application/json", forHTTPHeaderField: "Content-Type")
request.setValue(
Self.refreshTokenIdempotencyKey(credentials.refreshToken),
forHTTPHeaderField: "Idempotency-Key"
)
request.setValue(UUID().uuidString, forHTTPHeaderField: "X-Request-ID")
request.httpBody = try JSONEncoder().encode(Body(refreshToken: credentials.refreshToken))
do {
let refreshed = try await sendGrantRequest(request)
guard refreshed.grantId == credentials.grantId,
refreshed.scopes == credentials.scopes else {
try? await store.delete()
throw ManagedGatewayError.invalidGrant
}
try await store.save(refreshed)
return refreshed
} catch let error as ManagedGatewayError where error == .invalidGrant {
try? await store.delete()
throw error
}
}
private func sendGrantRequest(
_ request: URLRequest
) async throws -> ManagedGatewayGrantCredentials {
do {
let (data, response) = try await session.data(for: request)
guard let http = response as? HTTPURLResponse else {
throw LLMError.transport("non-HTTP response")
}
guard (200..<300).contains(http.statusCode) else {
throw ManagedGatewayHTTP.error(
data: data,
status: http.statusCode,
requestId: http.value(forHTTPHeaderField: "X-Request-ID")
)
}
do {
return try Self.gatewayDecoder()
.decode(ManagedGatewayGrantTokenResponse.self, from: data)
.credentials(receivedAt: now())
} catch {
throw LLMError.decoding(String(describing: error))
}
} catch is CancellationError {
throw LLMError.cancelled
} catch let error as URLError where error.code == .cancelled {
throw LLMError.cancelled
} catch let error as URLError where error.code == .timedOut {
throw ManagedGatewayError.timeout
} catch let error as ManagedGatewayError {
throw error
} catch let error as LLMError {
throw error
} catch {
throw LLMError.transport(String(describing: error))
}
}
private func endpoint(_ path: String) -> URL {
baseURL.appending(path: path)
}
static func refreshTokenIdempotencyKey(_ refreshToken: String) -> String {
let digest = SHA256.hash(data: Data(refreshToken.utf8))
return "gateway-refresh-v1-" + digest.map { String(format: "%02x", $0) }.joined()
}
private static func gatewayDecoder() -> JSONDecoder {
let decoder = JSONDecoder()
decoder.dateDecodingStrategy = .custom { decoder in
let value = try decoder.singleValueContainer().decode(String.self)
let fractional = ISO8601DateFormatter()
fractional.formatOptions = [.withInternetDateTime, .withFractionalSeconds]
if let date = fractional.date(from: value) {
return date
}
let standard = ISO8601DateFormatter()
standard.formatOptions = [.withInternetDateTime]
guard let date = standard.date(from: value) else {
throw DecodingError.dataCorruptedError(
in: try decoder.singleValueContainer(),
debugDescription: "Invalid ISO-8601 date"
)
}
return date
}
return decoder
}
}
enum ManagedGatewayHTTP {
static func error(
data: Data,
status: Int,
requestId: String?
) -> ManagedGatewayError {
let decoded = decodeError(from: data)
let code = decoded?.code ?? HTTPURLResponse.localizedString(forStatusCode: status)
let resolvedRequestId = decoded?.requestId ?? requestId
switch code.lowercased() {
case "insufficient_credits", "insufficient_balance", "credit_balance_insufficient":
return .insufficientCredits
case "unauthorized", "invalid_gateway_refresh", "gateway_grant_denied", "invalid_grant":
return .invalidGrant
default:
return .server(code: code, status: status, requestId: resolvedRequestId)
}
}
static func decodeError(from data: Data) -> ManagedGatewayErrorResponse? {
if let direct = try? JSONDecoder().decode(ManagedGatewayErrorResponse.self, from: data) {
return direct
}
guard let object = try? JSONSerialization.jsonObject(with: data) as? [String: Any],
let nested = object["error"] as? [String: Any],
let code = nested["code"] as? String,
let message = nested["message"] as? String else {
return nil
}
return ManagedGatewayErrorResponse(
code: code,
message: message,
requestId: nested["requestId"] as? String ?? ""
)
}
}
@@ -0,0 +1,106 @@
// GatewayGrantCredentialStore.swift
// OSGKeyboard · Shared
//
// Shared Keychain storage dedicated to scope-limited gateway grants. Account
// session credentials intentionally never enter this service or access group item.
import Foundation
import Security
public protocol GatewayGrantCredentialStore: Sendable {
func load() async throws -> ManagedGatewayGrantCredentials?
func save(_ credentials: ManagedGatewayGrantCredentials) async throws
func delete() async throws
}
public struct GatewayGrantKeychainStore: GatewayGrantCredentialStore, @unchecked Sendable {
public enum StoreError: Error, Equatable, Sendable {
case unexpectedStatus(OSStatus)
case invalidStoredValue
}
private static let service = "com.osgkeyboard.gateway-grant"
private static let account = "scope-limited.active"
public init() {}
public func load() async throws -> ManagedGatewayGrantCredentials? {
var query = Self.baseQuery
query[kSecReturnData as String] = true
query[kSecMatchLimit as String] = kSecMatchLimitOne
var result: CFTypeRef?
let status = SecItemCopyMatching(query as CFDictionary, &result)
switch status {
case errSecSuccess:
guard let data = result as? Data,
let value = try? Self.decoder.decode(
ManagedGatewayGrantCredentials.self,
from: data
) else {
throw StoreError.invalidStoredValue
}
return value
case errSecItemNotFound:
return nil
default:
throw StoreError.unexpectedStatus(status)
}
}
public func save(_ credentials: ManagedGatewayGrantCredentials) async throws {
let data = try Self.encoder.encode(credentials)
let updateStatus = SecItemUpdate(
Self.baseQuery as CFDictionary,
[kSecValueData as String: data] as CFDictionary
)
switch updateStatus {
case errSecSuccess:
return
case errSecItemNotFound:
var query = Self.baseQuery
query[kSecValueData as String] = data
// The extension can refresh after reboot without making account
// credentials readable; this item contains only the limited grant.
query[kSecAttrAccessible as String] = kSecAttrAccessibleAfterFirstUnlockThisDeviceOnly
let addStatus = SecItemAdd(query as CFDictionary, nil)
guard addStatus == errSecSuccess else {
throw StoreError.unexpectedStatus(addStatus)
}
default:
throw StoreError.unexpectedStatus(updateStatus)
}
}
public func delete() async throws {
let status = SecItemDelete(Self.baseQuery as CFDictionary)
guard status == errSecSuccess || status == errSecItemNotFound else {
throw StoreError.unexpectedStatus(status)
}
}
private static var baseQuery: [String: Any] {
var query: [String: Any] = [
kSecClass as String: kSecClassGenericPassword,
kSecAttrService as String: service,
kSecAttrAccount as String: account,
kSecAttrSynchronizable as String: kCFBooleanFalse!
]
#if os(macOS)
query[kSecUseDataProtectionKeychain as String] = true
#endif
return query
}
private static var encoder: JSONEncoder {
let encoder = JSONEncoder()
encoder.dateEncodingStrategy = .iso8601
return encoder
}
private static var decoder: JSONDecoder {
let decoder = JSONDecoder()
decoder.dateDecodingStrategy = .iso8601
return decoder
}
}
@@ -0,0 +1,146 @@
// ManagedGatewayModels.swift
// OSGKeyboard · Shared
//
// Scope-limited credentials and transport models for the managed gateway.
import Foundation
public enum ManagedGatewayCapability: String, Codable, CaseIterable, Sendable {
case polish
case assistant = "ai"
case agent
case asr
}
/// Stable task contract shared with the account gateway. The app identifies
/// user intent; the server remains authoritative for model and feature policy.
public enum ManagedGatewayTaskKind: String, Codable, CaseIterable, Sendable {
case dictationPolish = "dictation_polish"
case translation
case editLastInput = "edit_last_input"
case aiQuestion = "ai_question"
case clipboardTransform = "clipboard_transform"
case customSkill = "custom_skill"
case agentPlanning = "agent_planning"
}
public struct ManagedGatewayGrantCredentials: Codable, Equatable, Sendable {
public static let maximumAccessLifetime: TimeInterval = 5 * 60
public let grantId: String
public let scopes: Set<ManagedGatewayCapability>
public let accessToken: String
public let accessExpiresAt: Date
public let refreshToken: String
public let refreshExpiresAt: Date
public let receivedAt: Date
public init(
grantId: String,
scopes: Set<ManagedGatewayCapability>,
accessToken: String,
accessExpiresAt: Date,
refreshToken: String,
refreshExpiresAt: Date,
receivedAt: Date = Date()
) {
self.grantId = grantId
self.scopes = scopes
self.accessToken = accessToken
self.accessExpiresAt = accessExpiresAt
self.refreshToken = refreshToken
self.refreshExpiresAt = refreshExpiresAt
self.receivedAt = receivedAt
}
/// Never trust an unexpectedly long access expiry. The gateway contract
/// deliberately limits extension-readable bearer credentials to five minutes.
public var effectiveAccessExpiresAt: Date {
min(accessExpiresAt, receivedAt.addingTimeInterval(Self.maximumAccessLifetime))
}
public func hasUsableAccessToken(
for scope: ManagedGatewayCapability,
at now: Date = Date(),
refreshLeeway: TimeInterval = 30
) -> Bool {
scopes.contains(scope)
&& !accessToken.isEmpty
&& effectiveAccessExpiresAt.timeIntervalSince(now) > refreshLeeway
}
public func hasUsableRefreshToken(at now: Date = Date()) -> Bool {
!refreshToken.isEmpty && refreshExpiresAt > now
}
}
public enum ManagedGatewayError: Error, LocalizedError, Equatable, Sendable {
case missingGrant
case scopeNotGranted(ManagedGatewayCapability)
case invalidGrant
case insufficientCredits
case timeout
case server(code: String, status: Int, requestId: String?)
public var errorDescription: String? {
switch self {
case .missingGrant:
return SharedL10n.string("managed.error.grantUnavailable")
case .scopeNotGranted(let scope):
return SharedL10n.format(
"managed.error.scopeNotGranted",
language: nil,
scope.rawValue
)
case .invalidGrant:
return SharedL10n.string("managed.error.grantRejected")
case .insufficientCredits:
return SharedL10n.string("managed.error.insufficientCredits")
case .timeout:
return SharedL10n.string("managed.error.timeout")
case .server(let code, let status, _):
return SharedL10n.format(
"managed.error.server",
language: nil,
code,
status
)
}
}
}
struct ManagedGatewayGrantTokenResponse: Decodable {
let grantId: String
let scopes: Set<ManagedGatewayCapability>
let accessToken: String
let accessExpiresAt: Date
let refreshToken: String
let refreshExpiresAt: Date
func credentials(receivedAt: Date) -> ManagedGatewayGrantCredentials {
ManagedGatewayGrantCredentials(
grantId: grantId,
scopes: scopes,
accessToken: accessToken,
accessExpiresAt: accessExpiresAt,
refreshToken: refreshToken,
refreshExpiresAt: refreshExpiresAt,
receivedAt: receivedAt
)
}
}
struct ManagedGatewayErrorResponse: Decodable, Sendable {
let code: String
let message: String
let requestId: String
}
struct ManagedGatewayTextRequest: Encodable, Sendable {
let input: String
let context: String?
let maxOutputTokens: Int
let temperature: Double
let stream: Bool
let taskKind: ManagedGatewayTaskKind
}
@@ -0,0 +1,16 @@
// ManagedGatewayScopePolicy.swift
// OSGKeyboard · Shared
//
// Computes the smallest grant needed by the currently enabled managed paths.
import Foundation
public enum ManagedGatewayScopePolicy {
public static func scopes(engineMode: String) -> Set<ManagedGatewayCapability> {
var scopes: Set<ManagedGatewayCapability> = [.polish, .assistant, .agent]
if engineMode == "cloud" {
scopes.insert(.asr)
}
return scopes
}
}
@@ -0,0 +1,549 @@
// ManagedLLMClient.swift
// OSGKeyboard · Shared
//
// LLMClient implementation for scope-limited managed polish, AI and agent calls.
import Foundation
public struct ManagedLLMClient: LLMClient {
public enum Capability: String, Sendable {
case polish
case assistant = "ai"
case agent
var grantScope: ManagedGatewayCapability {
switch self {
case .polish: .polish
case .assistant: .assistant
case .agent: .agent
}
}
var defaultTaskKind: ManagedGatewayTaskKind {
switch self {
case .polish: .dictationPolish
case .assistant: .aiQuestion
case .agent: .agentPlanning
}
}
}
private struct Attempt {
let input: String
let context: String?
let timeout: TimeInterval?
let options: LLMGenerationOptions
let requestId: String
let forceRefresh: Bool
func forcingRefresh() -> Self {
Self(
input: input,
context: context,
timeout: timeout,
options: options,
requestId: requestId,
forceRefresh: true
)
}
}
public let capability: Capability
public let taskKind: ManagedGatewayTaskKind
public let requestTimeout: TimeInterval
private let baseURL: URL
private let grants: GatewayGrantCoordinator
private let session: URLSession
private let requestId: @Sendable () -> String
public init(
capability: Capability,
taskKind: ManagedGatewayTaskKind? = nil,
grants: GatewayGrantCoordinator,
baseURL: URL = GatewayGrantCoordinator.defaultBaseURL,
session: URLSession = .shared,
requestTimeout: TimeInterval = 15,
requestId: @escaping @Sendable () -> String = { UUID().uuidString }
) {
self.capability = capability
self.taskKind = taskKind ?? capability.defaultTaskKind
self.grants = grants
self.baseURL = baseURL
self.session = session
self.requestTimeout = requestTimeout
self.requestId = requestId
}
public func polish(
_ text: String,
systemPrompt: String,
timeout: TimeInterval?
) async throws -> String {
try await polish(
text,
systemPrompt: systemPrompt,
timeout: timeout,
options: .polishDefault
)
}
public func polish(
_ text: String,
systemPrompt: String,
timeout: TimeInterval?,
options: LLMGenerationOptions
) async throws -> String {
try await executeBuffered(
input: text,
context: systemPrompt.nilIfEmpty,
timeout: timeout,
options: options
)
}
public func complete(
messages: [LLMRequest.Message],
timeout: TimeInterval?,
options: LLMGenerationOptions
) async throws -> String {
let payload = Self.payload(from: messages)
return try await executeBuffered(
input: payload.input,
context: payload.context,
timeout: timeout,
options: options
)
}
public func completeStreaming(
messages: [LLMRequest.Message],
timeout: TimeInterval?,
options: LLMGenerationOptions
) -> AsyncThrowingStream<LLMStreamEvent, Error> {
AsyncThrowingStream { continuation in
let task = Task {
do {
let payload = Self.payload(from: messages)
let logicalRequestId = requestId()
let attempt = Attempt(
input: payload.input,
context: payload.context,
timeout: timeout,
options: options,
requestId: logicalRequestId,
forceRefresh: false
)
var emittedVisibleText = false
do {
try await streamAttempt(attempt) { chunk in
emittedVisibleText = true
continuation.yield(.delta(chunk))
}
} catch ManagedGatewayError.invalidGrant where !emittedVisibleText {
try await streamAttempt(attempt.forcingRefresh()) { chunk in
emittedVisibleText = true
continuation.yield(.delta(chunk))
}
}
continuation.finish()
} catch is CancellationError {
continuation.finish(throwing: LLMError.cancelled)
} catch let error as URLError where error.code == .cancelled {
continuation.finish(throwing: LLMError.cancelled)
} catch let error as URLError where error.code == .timedOut {
continuation.finish(throwing: ManagedGatewayError.timeout)
} catch let error as LLMError {
continuation.finish(throwing: error)
} catch let error as ManagedGatewayError {
continuation.finish(throwing: error)
} catch {
continuation.finish(throwing: LLMError.transport(String(describing: error)))
}
}
continuation.onTermination = { _ in task.cancel() }
}
}
private func executeBuffered(
input: String,
context: String?,
timeout: TimeInterval?,
options: LLMGenerationOptions
) async throws -> String {
let logicalRequestId = requestId()
let attempt = Attempt(
input: input,
context: context,
timeout: timeout,
options: options,
requestId: logicalRequestId,
forceRefresh: false
)
do {
return try await bufferedAttempt(attempt)
} catch ManagedGatewayError.invalidGrant {
do {
return try await bufferedAttempt(attempt.forcingRefresh())
} catch ManagedGatewayError.invalidGrant {
try? await grants.clearGrant()
throw ManagedGatewayError.invalidGrant
}
}
}
private func bufferedAttempt(_ attempt: Attempt) async throws -> String {
let request = try await makeRequest(
attempt,
stream: false
)
do {
let (data, response) = try await session.data(for: request)
guard let http = response as? HTTPURLResponse else {
throw LLMError.transport("non-HTTP response")
}
guard (200..<300).contains(http.statusCode) else {
throw ManagedGatewayHTTP.error(
data: data,
status: http.statusCode,
requestId: attempt.requestId
)
}
return try Self.responseText(from: data)
} catch is CancellationError {
throw LLMError.cancelled
} catch let error as URLError where error.code == .cancelled {
throw LLMError.cancelled
} catch let error as URLError where error.code == .timedOut {
throw ManagedGatewayError.timeout
} catch let error as ManagedGatewayError {
throw error
} catch let error as LLMError {
throw error
} catch {
throw LLMError.transport(String(describing: error))
}
}
private func streamAttempt(
_ attempt: Attempt,
onDelta: @escaping (String) -> Void
) async throws {
guard capability != .agent else {
// The server validates agent output as one structured response.
let text = try await bufferedAttempt(attempt)
if !text.isEmpty {
onDelta(text)
}
return
}
let request = try await makeRequest(
attempt,
stream: true
)
for try await payload in ManagedGatewayStreamTransport.payloads(
session: session,
request: request,
requestId: attempt.requestId
) {
try Task.checkCancellation()
if let error = Self.streamingError(from: payload, requestId: attempt.requestId) {
throw error
}
if let delta = Self.streamingDelta(from: payload), !delta.isEmpty {
onDelta(delta)
}
}
}
private func makeRequest(_ attempt: Attempt, stream: Bool) async throws -> URLRequest {
let trimmedInput = attempt.input.trimmingCharacters(in: .whitespacesAndNewlines)
guard !trimmedInput.isEmpty, trimmedInput.count <= 32_000 else {
throw ManagedGatewayError.server(
code: "invalid_request",
status: 400,
requestId: attempt.requestId
)
}
let boundedContext = attempt.context?
.trimmingCharacters(in: .whitespacesAndNewlines)
.nilIfEmpty
guard (boundedContext?.count ?? 0) <= 32_000 else {
throw ManagedGatewayError.server(
code: "invalid_request",
status: 400,
requestId: attempt.requestId
)
}
let token = try await grants.accessToken(
for: capability.grantScope,
forceRefresh: attempt.forceRefresh
)
let body = ManagedGatewayTextRequest(
input: trimmedInput,
context: boundedContext,
maxOutputTokens: min(max(attempt.options.maxTokens ?? 512, 1), 4_096),
temperature: min(max(attempt.options.temperature ?? 0.2, 0), 1),
stream: stream,
taskKind: taskKind
)
var request = URLRequest(
url: baseURL.appending(path: "v1/gateway/llm/\(capability.rawValue)")
)
request.httpMethod = "POST"
request.setValue("application/json", forHTTPHeaderField: "Content-Type")
request.setValue(stream ? "text/event-stream" : "application/json", forHTTPHeaderField: "Accept")
request.setValue("Bearer \(token)", forHTTPHeaderField: "Authorization")
request.setValue(attempt.requestId, forHTTPHeaderField: "X-Request-ID")
request.timeoutInterval = attempt.timeout ?? requestTimeout
request.httpBody = try JSONEncoder().encode(body)
return request
}
static func payload(
from messages: [LLMRequest.Message]
) -> (input: String, context: String?) {
guard let inputIndex = messages.lastIndex(where: { $0.role == "user" }) else {
return ("", Self.contextText(from: messages))
}
let input = messages[inputIndex].content
var contextMessages = messages
contextMessages.remove(at: inputIndex)
return (input, Self.contextText(from: contextMessages))
}
private static func contextText(from messages: [LLMRequest.Message]) -> String? {
messages
.filter { !$0.content.trimmingCharacters(in: .whitespacesAndNewlines).isEmpty }
.map { "\($0.role):\n\($0.content)" }
.joined(separator: "\n\n")
.nilIfEmpty
}
static func responseText(from data: Data) throws -> String {
let raw = String(data: data, encoding: .utf8)?
.trimmingCharacters(in: .whitespacesAndNewlines) ?? ""
guard !raw.isEmpty else { throw LLMError.decoding("empty managed gateway response") }
guard let json = try? JSONSerialization.jsonObject(with: data) else {
return raw
}
return extractedText(from: json)?.trimmingCharacters(in: .whitespacesAndNewlines)
.nilIfEmpty ?? raw
}
private static func extractedText(from value: Any) -> String? {
if let string = value as? String {
return string
}
if let array = value as? [Any] {
let values = array.compactMap(extractedText(from:))
return values.isEmpty ? nil : values.joined()
}
guard let object = value as? [String: Any] else { return nil }
for key in ["output_text", "text"] {
if let text = object[key] as? String, !text.isEmpty {
return text
}
}
if let content = object["content"] {
if let text = content as? String, !text.isEmpty {
return text
}
if let extracted = extractedText(from: content), !extracted.isEmpty {
return extracted
}
}
if let message = object["message"] as? [String: Any],
let extracted = extractedText(from: message) {
return extracted
}
if let choices = object["choices"] as? [[String: Any]],
let first = choices.first {
if let message = first["message"] as? [String: Any],
let extracted = extractedText(from: message) {
return extracted
}
if let text = first["text"] as? String {
return text
}
}
if let data = object["data"], let extracted = extractedText(from: data) {
return extracted
}
if let output = object["output"], let extracted = extractedText(from: output) {
return extracted
}
return nil
}
static func streamingDelta(from data: Data) -> String? {
if let value = LLMStreamDeltaParser.responsesOutputTextDelta(from: data)
?? LLMStreamDeltaParser.chatCompletionsDelta(from: data)
?? LLMStreamDeltaParser.anthropicTextDelta(from: data) {
return value
}
guard let json = try? JSONSerialization.jsonObject(with: data) as? [String: Any] else {
return String(data: data, encoding: .utf8)
}
for key in ["delta", "text", "content"] {
if let value = json[key] as? String, !value.isEmpty {
return value
}
}
return nil
}
private static func streamingError(
from data: Data,
requestId: String
) -> ManagedGatewayError? {
guard let object = try? JSONSerialization.jsonObject(with: data) as? [String: Any],
let code = object["code"] as? String,
object["message"] != nil || code.hasSuffix("_error") else {
return nil
}
if ["insufficient_credits", "insufficient_balance"].contains(code) {
return .insufficientCredits
}
if ["unauthorized", "gateway_grant_denied", "invalid_grant"].contains(code) {
return .invalidGrant
}
return .server(code: code, status: 200, requestId: requestId)
}
}
public enum ManagedGatewayLLMClientFactory {
public static func polish(
taskKind: ManagedGatewayTaskKind = .dictationPolish,
grants: GatewayGrantCoordinator,
baseURL: URL = GatewayGrantCoordinator.defaultBaseURL,
session: URLSession = .shared
) -> any LLMClient {
ManagedLLMClient(
capability: .polish,
taskKind: taskKind,
grants: grants,
baseURL: baseURL,
session: session
)
}
public static func ai(
taskKind: ManagedGatewayTaskKind = .aiQuestion,
grants: GatewayGrantCoordinator,
baseURL: URL = GatewayGrantCoordinator.defaultBaseURL,
session: URLSession = .shared
) -> any LLMClient {
ManagedLLMClient(
capability: .assistant,
taskKind: taskKind,
grants: grants,
baseURL: baseURL,
session: session
)
}
public static func agent(
taskKind: ManagedGatewayTaskKind = .agentPlanning,
grants: GatewayGrantCoordinator,
baseURL: URL = GatewayGrantCoordinator.defaultBaseURL,
session: URLSession = .shared
) -> any LLMClient {
ManagedLLMClient(
capability: .agent,
taskKind: taskKind,
grants: grants,
baseURL: baseURL,
session: session
)
}
}
private extension String {
var nilIfEmpty: String? {
isEmpty ? nil : self
}
}
enum ManagedGatewayStreamTransport {
static func payloads(
session: URLSession,
request: URLRequest,
requestId: String
) -> AsyncThrowingStream<Data, Error> {
AsyncThrowingStream { continuation in
let task = Task {
do {
let (bytes, response) = try await session.bytes(for: request)
guard let http = response as? HTTPURLResponse else {
throw LLMError.transport("non-HTTP response")
}
guard (200..<300).contains(http.statusCode) else {
var body = Data()
for try await byte in bytes {
body.append(byte)
}
throw ManagedGatewayHTTP.error(
data: body,
status: http.statusCode,
requestId: requestId
)
}
let isEventStream = http.value(forHTTPHeaderField: "Content-Type")?
.lowercased()
.contains("text/event-stream") == true
if !isEventStream {
var body = Data()
for try await byte in bytes {
try Task.checkCancellation()
body.append(byte)
}
if !body.isEmpty {
continuation.yield(body)
}
continuation.finish()
return
}
var line = Data()
for try await byte in bytes {
try Task.checkCancellation()
if byte == UInt8(ascii: "\n") {
yieldSSELine(line, to: continuation)
line.removeAll(keepingCapacity: true)
} else if byte != UInt8(ascii: "\r") {
line.append(byte)
}
}
yieldSSELine(line, to: continuation)
continuation.finish()
} catch is CancellationError {
continuation.finish(throwing: LLMError.cancelled)
} catch let error as URLError where error.code == .cancelled {
continuation.finish(throwing: LLMError.cancelled)
} catch let error as URLError where error.code == .timedOut {
continuation.finish(throwing: ManagedGatewayError.timeout)
} catch {
continuation.finish(throwing: error)
}
}
continuation.onTermination = { _ in task.cancel() }
}
}
private static func yieldSSELine(
_ line: Data,
to continuation: AsyncThrowingStream<Data, Error>.Continuation
) {
guard let payload = LLMStreamTransport.sseDataPayload(fromLineBytes: line),
payload != Data("[DONE]".utf8) else {
return
}
continuation.yield(payload)
}
}