Providers

Supported AI providers and how they integrate with CLI Proxy API

Overview

CLI Proxy API supports multiple AI providers, each with unique authentication methods, model catalogs, and capabilities.

Supported Providers

Gemini

Google’s Gemini models via OAuth, API keys, or Vertex AI

Claude

Anthropic Claude models via OAuth or API keys

Codex

OpenAI GPT models via OAuth

Qwen

Alibaba Qwen Code models via OAuth

iFlow

Z.ai GLM models via OAuth

Antigravity

Google’s code assistance platform

Kimi

Moonshot AI models via OAuth

AI Studio

Google AI Studio API keys

Vertex AI

Google Cloud Vertex AI endpoints

Provider Details

Gemini CLI

Authentication: OAuth 2.0

Models:

• gemini-2.5-pro - Latest Pro model
• gemini-2.5-flash - Fast Flash model
• gemini-3-pro-preview - Preview models
• gemini-3-pro-high - High performance variant

Configuration:

Authentication

cli-proxy-api auth gemini

# Optional: Configure OAuth model aliases
oauth-model-alias:
  gemini-cli:
    - name: "gemini-2.5-pro"
      alias: "g2.5p"
      fork: true  # Keep both original and alias

# Optional: Exclude models
oauth-excluded-models:
  gemini-cli:
    - "*-preview"     # Exclude all preview models
    - "gemini-2.5-*"  # Exclude 2.5 series

POST http://localhost:8317/v1/chat/completions
Authorization: Bearer your-api-key
Content-Type: application/json

{
  "model": "gemini-2.5-pro",
  "messages": [
    {"role": "user", "content": "Hello!"}
  ]
}

Features:

• Multi-account load balancing
• Automatic token refresh
• Virtual parent grouping for project-based accounts
• Support for thinking models with budget configuration

File Format:

~/.cli-proxy-api/[e-mail]

AI Studio (Gemini API Keys)

Authentication: API Key

Configuration:

config.yaml

gemini-api-key:
  - api-key: "AIzaSyDemoKey123456789"
    prefix: "studio"  # Optional: target with "studio/model-name"
    base-url: "https://generativelanguage.googleapis.com"
    headers:
      X-Custom-Header: "custom-value"
    proxy-url: "socks5://proxy.example.com:1080"
    models:
      - name: "gemini-2.5-flash"
        alias: "gemini-flash"
    excluded-models:
      - "gemini-2.5-pro"
      - "*-preview"

Features:

• Direct API key access
• Custom base URLs for relay services
• Per-key proxy configuration
• Model aliasing and exclusion

Vertex AI

Authentication: API Key (Vertex-compatible endpoints)

Configuration:

config.yaml

vertex-api-key:
  - api-key: "vk-123456789"  # x-goog-api-key header
    prefix: "vertex"
    base-url: "https://example.com/api"  # e.g. https://zenmux.ai/api
    proxy-url: "socks5://proxy.example.com:1080"
    headers:
      X-Custom-Header: "custom-value"
    models:
      - name: "gemini-2.5-flash"
        alias: "vertex-flash"
      - name: "gemini-2.5-pro"
        alias: "vertex-pro"
    excluded-models:
      - "imagen-3.0-generate-002"
      - "imagen-*"

Use Case: Third-party Vertex AI endpoints that use API keys instead of service accounts.

Claude Code

Authentication: OAuth 2.0

Models:

• claude-sonnet-4 - Claude Sonnet 4
• claude-opus-4-5-20251101 - Claude Opus 4.5
• claude-3-5-sonnet-20241022 - Claude 3.5 Sonnet
• claude-3-5-haiku-20241022 - Claude 3.5 Haiku

Configuration:

Authentication

cli-proxy-api auth claude

# Optional: Configure OAuth model aliases
oauth-model-alias:
  claude:
    - name: "claude-sonnet-4-5-20250929"
      alias: "cs4.5"

# Optional: Exclude models
oauth-excluded-models:
  claude:
    - "claude-3-5-haiku-20241022"
    - "*-thinking"

Features:

• Multi-organization support
• Custom TLS fingerprinting (bypasses Cloudflare)
• Request cloaking for non-Claude-Code clients
• Automatic session management

File Format:

~/.cli-proxy-api/claude_oauth_org-uuid.json

Request Cloaking:

Clause Code has detection mechanisms to identify non-official clients. CLI Proxy API supports request cloaking:

config.yaml

claude-api-key:
  - api-key: "sk-ant-api03-..."
    cloak:
      mode: "auto"           # "auto", "always", or "never"
      strict-mode: false     # Strip user system messages
      sensitive-words:       # Obfuscate with zero-width chars
        - "API"
        - "proxy"
      cache-user-id: true    # Reuse cached user_id

Cloaking features:

• auto: Cloak only for non-Claude-Code clients
• always: Always apply cloaking
• never: Disable cloaking
• strict-mode: Strip all user system messages
• sensitive-words: Obfuscate specific words

Claude API (API Keys)

Authentication: API Key

Configuration:

config.yaml

claude-api-key:
  - api-key: "sk-ant-api03-..."
    prefix: "official"  # Optional
    base-url: "https://api.anthropic.com"  # Official endpoint
    headers:
      X-Custom-Header: "custom-value"
    proxy-url: "socks5://proxy.example.com:1080"
    models:
      - name: "claude-3-5-sonnet-20241022"
        alias: "claude-sonnet-latest"
    excluded-models:
      - "claude-opus-4-5-20251101"
      - "*-haiku*"

Use Case: Official Anthropic API keys or relay services.

OpenAI Codex

Authentication: OAuth 2.0

Models:

• gpt-5-codex - GPT-5 for coding
• gpt-5 - GPT-5 base model
• o3-mini - O3 mini model
• o1-preview - O1 preview model

Configuration:

Authentication

cli-proxy-api auth codex

# Optional: Configure OAuth model aliases
oauth-model-alias:
  codex:
    - name: "gpt-5"
      alias: "g5"

# Optional: Exclude models
oauth-excluded-models:
  codex:
    - "gpt-5-codex-mini"
    - "*-mini"

Features:

• JWT-based session tokens
• Organization membership tracking
• Refresh token rotation
• Support for reasoning models

File Format:

~/.cli-proxy-api/[e-mail]

Codex API (API Keys)

Authentication: API Key

Configuration:

config.yaml

codex-api-key:
  - api-key: "sk-atSM..."
    prefix: "openai"
    base-url: "https://api.openai.com/v1"  # Or custom endpoint
    headers:
      X-Custom-Header: "custom-value"
    proxy-url: "socks5://proxy.example.com:1080"
    models:
      - name: "gpt-5-codex"
        alias: "codex-latest"
    excluded-models:
      - "gpt-5.1"
      - "*-mini"

Use Case: OpenAI API keys or OpenAI-compatible relay services.

Qwen Code

Authentication: OAuth 2.0

Models:

• qwen3-coder-plus - Qwen 3 Coder Plus
• qwen3.5-plus - Qwen 3.5 Plus
• qwen-max - Qwen Max model

Configuration:

Authentication

cli-proxy-api auth qwen

oauth-model-alias:
  qwen:
    - name: "qwen3-coder-plus"
      alias: "qwen-plus"

oauth-excluded-models:
  qwen:
    - "vision-model"

Features:

• Alibaba Cloud integration
• Multi-account support
• Cookie-based authentication

File Format:

~/.cli-proxy-api/qwen_oauth_account-id.json

iFlow (Z.ai GLM)

Authentication: OAuth 2.0

Models:

• glm-4.7 - GLM 4.7
• glm-5 - GLM 5 (Pro users)
• tstars2.0 - TStar models

Configuration:

Authentication

cli-proxy-api auth iflow

oauth-model-alias:
  iflow:
    - name: "glm-4.7"
      alias: "glm-god"

oauth-excluded-models:
  iflow:
    - "tstars2.0"

Sponsor: This project is sponsored by Z.ai’s GLM CODING PLAN starting at $10/month.

Features:

• GLM CODING PLAN integration
• Multi-model support
• Cookie-based session management

File Format:

~/.cli-proxy-api/iflow_oauth_account-id.json

Antigravity

Authentication: OAuth 2.0 (Google OAuth)

Models:

• gemini-3-pro-high - High performance model
• gemini-3-pro-preview - Preview models
• Other Gemini variants

Configuration:

Authentication

cli-proxy-api auth antigravity

oauth-model-alias:
  antigravity:
    - name: "gemini-3-pro-high"
      alias: "gemini-3-pro-preview"

oauth-excluded-models:
  antigravity:
    - "gemini-3-pro-preview"

Features:

• Google’s code assistance platform
• Project ID management
• Onboarding flow for new users
• Tier-based access

File Format:

~/.cli-proxy-api/[e-mail]

Kimi

Authentication: OAuth 2.0

Models:

• kimi-k2.5 - Kimi K2.5
• kimi-k2 - Kimi K2
• kimi-k2-thinking - Thinking variant

Configuration:

Authentication

cli-proxy-api auth kimi

oauth-model-alias:
  kimi:
    - name: "kimi-k2.5"
      alias: "k2.5"

oauth-excluded-models:
  kimi:
    - "kimi-k2-thinking"

Features:

• Moonshot AI integration
• Cookie-based authentication
• Multi-account support

File Format:

~/.cli-proxy-api/kimi_oauth_account-id.json

OpenAI Compatibility

Authentication: API Key

Use Case: Any OpenAI-compatible endpoint (OpenRouter, local models, custom APIs)

Configuration:

config.yaml

openai-compatibility:
  - name: "openrouter"
    prefix: "router"  # Use as "router/model-name"
    base-url: "https://openrouter.ai/api/v1"
    headers:
      HTTP-Referer: "https://yourapp.com"
      X-Title: "Your App"
    api-key-entries:
      - api-key: "sk-or-v1-abc123"
        proxy-url: "socks5://proxy.example.com:1080"
      - api-key: "sk-or-v1-xyz789"  # Multiple keys for load balancing
    models:
      - name: "anthropic/claude-3.5-sonnet"
        alias: "claude-sonnet"
      - name: "google/gemini-pro"
        alias: "gemini-pro"
      # Internal model pools: same alias, multiple upstream models
      - name: "qwen3.5-plus"
        alias: "claude-opus-4.66"
      - name: "glm-5"
        alias: "claude-opus-4.66"
      - name: "kimi-k2.5"
        alias: "claude-opus-4.66"

Model Pools:

You can map multiple upstream models to the same alias for automatic failover:

models:
  - name: "qwen3.5-plus"
    alias: "best-model"
  - name: "glm-5"
    alias: "best-model"
  - name: "claude-3.5-sonnet"
    alias: "best-model"

Requests to best-model will round-robin across qwen3.5-plus, glm-5, and claude-3.5-sonnet. If one fails before producing output, the next model in the pool is tried.

Provider Comparison

Provider	Auth Method	Multi-Account	Streaming	Thinking	Image Input
Gemini CLI	OAuth	✅	✅	✅	✅
AI Studio	API Key	✅	✅	✅	✅
Vertex AI	API Key	✅	✅	✅	✅
Claude Code	OAuth	✅	✅	✅	✅
Claude API	API Key	✅	✅	✅	✅
Codex	OAuth	✅	✅	✅	❌
Codex API	API Key	✅	✅	✅	❌
Qwen	OAuth	✅	✅	❌	✅
iFlow	OAuth	✅	✅	❌	❌
Antigravity	OAuth	✅	✅	✅	✅
Kimi	OAuth	✅	✅	✅	❌
OpenAI Compat	API Key	✅	✅	Varies	Varies

Provider Selection

When a request specifies a model, CLI Proxy API:

1. Resolves aliases - Converts client model names to upstream names
2. Finds providers - Identifies which providers offer the model
3. Filters credentials - Excludes disabled/cooldown/quota-exceeded credentials
4. Applies routing - Uses round-robin or fill-first strategy
5. Executes request - Sends to selected provider

Model Prefix Routing

Force a specific provider using prefixes:

# Force personal Gemini account
curl -d '{"model": "personal/gemini-2.5-pro", ...}' ...

# Force work Claude account
curl -d '{"model": "work/claude-sonnet-4", ...}' ...

# Force OpenRouter provider
curl -d '{"model": "router/anthropic/claude-3.5-sonnet", ...}' ...

See Routing for details.

Provider-Specific Features

Thinking Models

Providers that support thinking/reasoning:

Gemini Thinking Budget

{
  "model": "gemini-2.5-pro",
  "messages": [...],
  "generationConfig": {
    "thinkingConfig": {
      "thinkingBudget": 32768  // Token budget for thinking
    }
  }
}

{
  "model": "gpt-5",
  "messages": [...],
  "reasoning": {
    "effort": "high"  // "low", "medium", "high"
  }
}

Multimodal Input

Providers supporting image input:

{
  "model": "gemini-2.5-pro",
  "messages": [
    {
      "role": "user",
      "content": [
        {"type": "text", "text": "What's in this image?"},
        {
          "type": "image_url",
          "image_url": {
            "url": "data:image/jpeg;base64,/9j/4AAQ..."
          }
        }
      ]
    }
  ]
}

WebSocket API

Some providers support WebSocket connections for lower latency:

config.yaml

ws-auth: true  # Enable WebSocket authentication

Auth File

{
  "access_token": "...",
  "attributes": {
    "websockets": "true"
  }
}

Connect via:

ws://localhost:8317/v1/ws

Adding Custom Providers

See the Custom Provider Example for implementing custom provider integrations using the SDK.

Next Steps

OAuth Setup

Authenticate with OAuth providers

API Keys

Configure API key providers

Routing

Learn about provider selection

Model Mappings

Configure model aliases and mappings