Codex Integration

Connect the Codex desktop app to GodeX by adding a custom provider in ~/.codex/config.toml:

model = "gpt-5.5"
model_provider = "godex"

[model_providers.godex]
name = "GodeX"
base_url = "http://127.0.0.1:5678/v1"
wire_api = "responses"
requires_openai_auth = false
supports_websockets = false

How It Works

Codex desktop app
    │
    │  Responses API
    │  POST /v1/responses
    ▼
GodeX (localhost:5678)
    │
    │  Chat Completions API
    ▼
DeepSeek · Zhipu · MiniMax · Xiaomi MiMo

Codex speaks the OpenAI Responses protocol. GodeX sits in the middle and translates to each provider's Chat Completions protocol. To Codex, GodeX is an ordinary OpenAI-compatible endpoint.

Key Fields

Field	Value	Notes
model	"gpt-5.5"	Codex model alias; resolved by GodeX models.aliases
model_provider	"godex"	Points to the [model_providers.godex] block
base_url	"http://127.0.0.1:5678/v1"	GodeX server address
wire_api	"responses"	Must be "responses" — GodeX speaks Responses API
requires_openai_auth	false	GodeX does not require OpenAI authentication
supports_websockets	false	GodeX does not support WebSocket transport

Available Model Aliases

GodeX ships with these Codex model aliases in godex.yaml:

Codex Model	Purpose	Routes To
gpt-5.5	Default: complex coding, computer use, research	deepseek/deepseek-v4-pro
gpt-5.4	Flagship: coding + reasoning + tool use	deepseek/deepseek-v4-pro
gpt-5.4-mini	Sub-agent tasks	zhipu/glm-5.1
gpt-5.3-codex	Codex coding: complex software engineering	deepseek/deepseek-v4-pro
gpt-5.3-codex-spark	Near-real-time coding iteration	zhipu/glm-5.1

Model aliases are managed entirely in godex.yaml. Codex only needs the alias name — no client-side changes required when switching providers.

Verification

Health Check

curl http://localhost:5678/health

List Models

curl http://localhost:5678/v1/models

Test Request

curl http://localhost:5678/v1/responses \
  -H 'content-type: application/json' \
  -d '{"model":"gpt-5.5","input":"Hello, who are you?"}'

Reasoning Effort

Codex model_reasoning_effort (low / medium / high / xhigh) is bridged by GodeX per provider:

Provider	Behavior
DeepSeek	Native reasoning_effort, passed through directly
Zhipu	Boolean thinking switch, mapped from effort level
MiniMax	No native reasoning, ignored
Xiaomi MiMo	Boolean thinking switch, similar to Zhipu

GodeX includes compatibility diagnostics in responses, so you always know what was degraded or ignored.