API - LLM

LLMProvider 接口

interface LLMProvider {
  chat(messages: MessageData[], options?: LLMOptions): Promise<LLMResponse>
  chatStream(messages: MessageData[], options?: LLMOptions): AsyncIterable<LLMStreamEvent>
  getTokenCount(messages: MessageData[]): number
}

---

LLMResponse

interface LLMResponse {
  content: string
  model: string
  usage?: {
    promptTokens: number
    completionTokens: number
    totalTokens: number
  }
  latencyMs?: number
  errorCode?: LLMResponseErrorCode
}

enum LLMResponseErrorCode {
  SUCCESS
  TIMEOUT
  CONNECTION_ERROR
  RATE_LIMITED
  SERVER_ERROR
  AUTH_ERROR
  BAD_REQUEST
  ABORT
  UNKNOWN
}

---

OpenAIProvider

import { OpenAIProvider } from 'kagent-ts'

new OpenAIProvider(config: OpenAIConfig)

interface OpenAIConfig {
  apiKey: string
  model?: string                    // 默认: "gpt-4o"
  baseURL?: string                  // 默认: "https://api.openai.com/v1"
  timeout?: number                  // 默认: 60000
  maxRetries?: number               // 默认: 3
  temperature?: number
  maxTokens?: number
  topP?: number
  retryConfig?: OpenAIRetryConfig
}

interface OpenAIRetryConfig {
  maxRetries?: number               // 默认: 3
  initialBackoffMs?: number         // 默认: 1000
  maxBackoffMs?: number             // 默认: 30000
  backoffMultiplier?: number       // 默认: 2
}

---

AnthropicProvider

import { AnthropicProvider } from 'kagent-ts'

new AnthropicProvider(config: AnthropicConfig)

interface AnthropicConfig {
  apiKey: string
  model?: string                    // 默认: "claude-sonnet-4-6"
  baseURL?: string
  timeout?: number                  // 默认: 60000
  maxRetries?: number               // 默认: 3
  maxTokens?: number
  cacheSystemPrompt?: boolean       // 默认: false
}

---

FallbackProvider

import { FallbackProvider } from 'kagent-ts'

new FallbackProvider(config: FallbackProviderConfig)

interface FallbackProviderConfig {
  providers: LLMProvider[]
  maxRetriesPerProvider?: number    // 默认: 2
}

---

RateLimitedProvider

import { RateLimitedProvider } from 'kagent-ts'

new RateLimitedProvider(config: RateLimitedProviderConfig)

interface RateLimitedProviderConfig {
  provider: LLMProvider
  maxCallsPerMinute: number
  windowSizeMs?: number             // 默认: 60000
}

---

ModelRouter

import { ModelRouter } from 'kagent-ts'

new ModelRouter(config: ModelRouterConfig)

interface ModelRouterConfig {
  routes: {
    main: LLMProvider
    subAgent?: LLMProvider
    reflection?: LLMProvider
    lightweight?: LLMProvider
  }
}

// 方法
router.forSubAgent(): LLMProvider
router.forReflection(): LLMProvider
router.forLightweight(): LLMProvider

---

createLLMProvider

import { createLLMProvider } from 'kagent-ts'

createLLMProvider(config: LLMProviderConfig): LLMProvider

interface LLMProviderConfig {
  apiKey: string
  model: string
  baseURL?: string
  provider?: "openai" | "anthropic" | "auto"  // 默认: "auto"
}

---

TokenBudget

import { TokenBudget } from 'kagent-ts'

new TokenBudget(config: TokenBudgetConfig)

interface TokenBudgetConfig {
  maxTokens: number
  warningThreshold?: number
}

interface TokenBudgetStatus {
  used: number
  remaining: number
  maxTokens: number
  warningThreshold: number
  isWarning: boolean
  isExhausted: boolean
}

---

LLMNetworkError

import { LLMNetworkError } from 'kagent-ts'

class LLMNetworkError extends Error {
  cause: NetworkErrorCause
  statusCode?: number
}

type NetworkErrorCause =
  | 'timeout'
  | 'connection'
  | 'rate_limited'
  | 'server_error'
  | 'abort'
  | 'dns'
  | 'tls'

下一步

• API - Tools — Tool 系统 API
• API - Messages — Message 类型 API
• API - Session — Session API