deepseek-proxy 0.2.1

Translate OpenAI Responses API ↔ DeepSeek Chat Completions API for codex

DeepSeek Proxy

Translate OpenAI Responses API (/v1/responses) requests into DeepSeek Chat Completions API calls — so codex (Claude Code CLI) can use DeepSeek as its backend model.

---

Install

pip install deepseek-proxy

Usage

# Set your API key
export DEEPSEEK_API_KEY=sk-your-key-here

# Start proxy (foreground)
deepseek-proxy

# Or start as a daemon (background)
deepseek-proxy start --daemon

# Stop the daemon
deepseek-proxy stop

# Restart
deepseek-proxy restart

# Custom host/port
deepseek-proxy --host 0.0.0.0 --port 8080 start --daemon

The proxy listens on http://127.0.0.1:8787 by default.

Configure codex

Point codex to the proxy via ~/.codex/config.toml:

Add to ~/.codex/config.toml:

model = "deepseek-v4-flash"
model_provider = "deepseek"

[model_providers.deepseek]
name = "DeepSeek"
base_url = "http://127.0.0.1:8787/v1"
env_key = "DEEPSEEK_API_KEY"
wire_api = "responses"

Field reference:

Field	Value	Description
model	"deepseek-v4-flash"	Default model for codex
model_provider	"deepseek"	Matches the provider section name
base_url	"http://127.0.0.1:8787/v1"	Proxy endpoint (no /responses suffix)
env_key	"DEEPSEEK_API_KEY"	Env var for the API key
wire_api	"responses"	Required value for this proxy

Verify

curl -X POST http://127.0.0.1:8787/v1/responses \
  -H "Authorization: Bearer $DEEPSEEK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"input":[{"role":"user","content":[{"type":"input_text","text":"hello"}]}],"model":"deepseek-v4-flash"}'

Why?

codex uses the OpenAI Responses API natively. DeepSeek provides a Chat Completions API. The two protocols are incompatible — this proxy bridges the gap.

---

GitHub