Skip to main content
Codebuff supports multiple AI model providers through OpenRouter, giving you access to the latest models from Anthropic, OpenAI, Google, and more.

​
OpenRouter Integration

Codebuff routes requests through OpenRouter by default, which provides access to a wide variety of models. The framework automatically handles:
  • Model routing and fallbacks
  • Cost tracking and optimization
  • Provider selection based on availability
  • Authentication and API key management

​
How It Works

When you make a request, Codebuff:
  1. Receives your API key via the Authorization header
  2. Routes the request to the Codebuff backend
  3. Forwards to OpenRouter with proper authentication
  4. Tracks usage and costs automatically
From sdk/src/impl/model-provider.ts:298-320: 
function createCodebuffBackendModel(
  apiKey: string,
  model: string,
): LanguageModel {
  return new OpenAICompatibleChatLanguageModel(model, {
    provider: 'codebuff',
    url: ({ path: endpoint }) =>
      new URL(path.join('/api/v1', endpoint), WEBSITE_URL).toString(),
    headers: () => ({
      Authorization: `Bearer ${apiKey}`,
      'user-agent': `ai-sdk/openai-compatible/VERSION/codebuff`,
    }),
    // ... metadata extraction for cost tracking
  })
}

​
Available Models

Codebuff provides access to leading models through OpenRouter. Here are the most commonly used:

​
Claude Models (Anthropic)

const claudeModels = {
  // Claude 4 (Latest)
  'anthropic/claude-sonnet-4.5': 'Claude Sonnet 4.5',
  'anthropic/claude-4-sonnet-20250522': 'Claude 4 Sonnet',
  'anthropic/claude-opus-4.1': 'Claude Opus 4.1',
  
  // Claude 3.5
  'anthropic/claude-3.5-haiku-20241022': 'Claude 3.5 Haiku',
  'anthropic/claude-3.5-sonnet-20240620': 'Claude 3.5 Sonnet',
}

​
OpenAI Models

const openaiModels = {
  'openai/gpt-4o-2024-11-20': 'GPT-4o',
  'openai/gpt-4o-mini-2024-07-18': 'GPT-4o Mini',
  'openai/gpt-5.1': 'GPT-5',
  'openai/o3-mini-2025-01-31': 'O3 Mini',
  'openai/gpt-4.1-nano': 'GPT-4.1 Nano',
}

​
Google Models

const googleModels = {
  'google/gemini-2.5-pro': 'Gemini 2.5 Pro',
  'google/gemini-2.5-flash': 'Gemini 2.5 Flash',
  'google/gemini-2.5-flash-preview:thinking': 'Gemini 2.5 Flash Thinking',
}

​
Other Providers

const otherModels = {
  'x-ai/grok-4-07-09': 'Grok 4',
}

​
Model Selection in Agents

When creating an agent, specify the model using OpenRouter’s format: 
import { agent } from '@codebuff/sdk'

export default agent({
  name: 'my-agent',
  model: 'anthropic/claude-4-sonnet-20250522',
  system: 'You are a helpful assistant.',
  tools: ['bash', 'read', 'edit'],
})

​
Model Aliases

Codebuff supports convenient aliases for common models: 
const modelAliases = {
  'opus-4': 'anthropic/claude-opus-4.1',
  'sonnet-4.5': 'anthropic/claude-sonnet-4.5',
  'sonnet-4': 'anthropic/claude-4-sonnet-20250522',
  'flash-2.5': 'google/gemini-2.5-flash',
  'gemini-2.5-pro': 'google/gemini-2.5-pro',
}
Use aliases in your agent definition: 
export default agent({
  name: 'my-agent',
  model: 'sonnet-4', // Resolves to anthropic/claude-4-sonnet-20250522
  // ...
})

​
Model Configuration

​
Provider Routing

Codebuff automatically configures provider routing for optimal performance and reliability: 
// From sdk/src/impl/llm.ts:37-49
const providerOrder = {
  'anthropic/claude-4-sonnet-20250522': [
    'Google',
    'Anthropic',
    'Amazon Bedrock',
  ],
  'anthropic/claude-sonnet-4.5': [
    'Google',
    'Anthropic',
    'Amazon Bedrock',
  ],
  'anthropic/claude-opus-4.1': ['Google', 'Anthropic'],
}

​
Custom Provider Options

For advanced use cases, configure provider routing in your agent: 
export default agent({
  name: 'my-agent',
  model: 'anthropic/claude-4-sonnet-20250522',
  providerOptions: {
    order: ['Anthropic', 'Google'],
    allow_fallbacks: true,
  },
})
See OpenRouter Provider Routing for details.

​
Claude OAuth (Advanced)

For Claude models, Codebuff supports direct OAuth authentication with Anthropic, bypassing OpenRouter:

​
Benefits

  • Direct API access to Anthropic
  • Potentially lower latency
  • Use your Claude subscription directly
  • No OpenRouter intermediary

​
How It Works

// From sdk/src/impl/model-provider.ts:171-194
export async function getModelForRequest(params: ModelRequestParams): Promise<ModelResult> {
  const { apiKey, model, skipClaudeOAuth } = params

  // Check if we should use Claude OAuth direct
  if (CLAUDE_OAUTH_ENABLED && !skipClaudeOAuth && !isClaudeOAuthRateLimited() && isClaudeModel(model)) {
    const claudeOAuthCredentials = await getValidClaudeOAuthCredentials()
    if (claudeOAuthCredentials) {
      return {
        model: createAnthropicOAuthModel(model, claudeOAuthCredentials.accessToken),
        isClaudeOAuth: true,
      }
    }
  }

  // Default: use Codebuff backend
  return {
    model: createCodebuffBackendModel(apiKey, model),
    isClaudeOAuth: false,
  }
}

​
Rate Limiting

ClaudeCoduff automatically handles Claude OAuth rate limits:
  • Detects 429 rate limit errors
  • Fetches reset time from Anthropic API
  • Automatically falls back to OpenRouter
  • Resumes OAuth when rate limit expires
// Rate limit tracking
let claudeOAuthRateLimitedUntil: number | null = null

export function markClaudeOAuthRateLimited(resetAt?: Date): void {
  const fiveMinutesFromNow = Date.now() + 5 * 60 * 1000
  claudeOAuthRateLimitedUntil = resetAt ? resetAt.getTime() : fiveMinutesFromNow
}

​
Cost Tracking

Codebuff tracks model usage costs automatically: 
type OpenRouterUsageAccounting = {
  cost: number | null
  costDetails: {
    upstreamInferenceCost: number | null
  }
}
Access cost information from the agent runtime metadata.

​
Next Steps

Build docs developers (and LLMs) love

Get started for freeTalk to us