DeepSeek Setup

Get API Key

1. Visit DeepSeek Platform
2. Sign up and log in
3. Go to API Keys page
4. Create a new API Key

How It Works

CCX supports multiple protocol endpoints for DeepSeek:

Claude Code CLI       ──→  /v1/messages          ──→  CCX  ──→  DeepSeek Anthropic endpoint
Codex CLI/App         ──→  /v1/responses         ──→  CCX  ──→  DeepSeek Chat endpoint
OpenAI-compatible app ──→  /v1/chat/completions  ──→  CCX  ──→  DeepSeek Chat endpoint

A single CCX instance serves all paths simultaneously.

Scenario 1: OpenAI Chat Protocol (General)

For any tool compatible with OpenAI Chat protocol.

Configuration

1. Go to CCX admin console, select Chat endpoint
2. Click "Add Channel"
3. Fill in:

Field	Value
Service Type	OpenAI Chat
Name	DeepSeek Chat
Base URL	https://api.deepseek.com
API Keys	Your DeepSeek API Key
Models	deepseek-v4-pro, deepseek-v4-flash

Client Configuration

export OPENAI_API_KEY="your-ccx-proxy-key"
export OPENAI_BASE_URL="http://localhost:3000/v1"

---

Scenario 2: Claude Code CLI

Claude Code CLI uses the Messages API. Configure a Claude service type channel pointing to DeepSeek's Anthropic-compatible endpoint.

Configuration

1. Go to CCX admin console, select Messages endpoint
2. Click "Add Channel"
3. Fill in:

Field	Value
Service Type	Claude
Name	DeepSeek Claude
Base URL	https://api.deepseek.com/anthropic
API Keys	Your DeepSeek API Key
Models	deepseek-v4-pro, deepseek-v4-flash

Model Mapping (Recommended)

Claude Code CLI sends requests with Claude model names (e.g., claude-opus-4-7). Configure model mapping for automatic redirection:

Request Model	Redirects To
opus	deepseek-v4-pro
sonnet	deepseek-v4-pro
haiku	deepseek-v4-flash

Client Configuration

export ANTHROPIC_API_KEY="your-ccx-proxy-key"
export ANTHROPIC_BASE_URL="http://localhost:3000"

WARNING

ANTHROPIC_BASE_URL should point to the CCX gateway root — do not append /v1 or /v1/messages.

---

Scenario 3: Codex CLI / App

Codex CLI uses the OpenAI Responses API. Configure a Chat service type channel under the Responses endpoint.

Configuration

1. Go to CCX admin console, select Responses endpoint
2. Click "Add Channel"
3. Fill in:

Field	Value
Service Type	OpenAI Chat
Name	DeepSeek Chat
Base URL	https://api.deepseek.com
API Keys	Your DeepSeek API Key
Models	deepseek-v4-pro, deepseek-v4-flash

4. After saving, edit the channel and enable Normalize Non-standard Chat Roles

Why enable this?

After Codex Responses requests are converted to Chat Completions, they may contain roles like developer that DeepSeek doesn't support. This option normalizes them to user before sending upstream.

Model Mapping (Recommended)

Request Model	Redirects To
gpt	deepseek-v4-pro
mini	deepseek-v4-flash

CCX uses the longest matching key first. gpt matches gpt-5, while mini matches gpt-5-mini.

Client Configuration

Codex CLI:

export OPENAI_API_KEY="your-ccx-proxy-key"
export OPENAI_BASE_URL="http://localhost:3000/v1"
codex "hello"

Codex App (VS Code / JetBrains):

Setting	Value
API Key	your-ccx-proxy-key
Base URL	http://localhost:3000/v1
Model	gpt-5 (CCX redirects to deepseek-v4-pro)

---

Available Models

Model	Description
deepseek-v4-pro	DeepSeek-V4 Pro flagship model
deepseek-v4-flash	DeepSeek-V4 Flash fast model
deepseek-chat	DeepSeek-V3 general chat (legacy)
deepseek-reasoner	DeepSeek-R1 reasoning model

Verify Configuration

curl http://localhost:3000/v1/models \
  -H "Authorization: Bearer your-ccx-proxy-key"

Troubleshooting

Issue	Solution
401 Unauthorized	Verify the key matches CCX's PROXY_ACCESS_KEY
Model not found	Check model names in channel config
Connection refused	Confirm CCX is running and Base URL is correct
Channel unhealthy	Check DeepSeek API Key and network access to api.deepseek.com
Claude Code format error	Ensure ANTHROPIC_BASE_URL points to root, not /v1