GitHub
⚡PURE PYTHON • ZERO DEPENDENCIES

Use Codex with
any model backend.

Route Codex IDE / CLI traffic to Ollama, NVIDIA Build, OpenRouter, vLLM, SGLang, LM Studio, DeepSeek, or any OpenAI-compatible API through one lightweight proxy.

Get Started →View GitHub
Near-zero overhead
Native Responses API
Cross-platform
codex-proxy

How it works

Codex IDE / CLI
Your local tools
Responses API
Standard format
Codex Proxy
Route & manage
Upstreams
Chat / Completions

Supported Providers

Works with local and remote providers. All speak OpenAI-compatible APIs. Codex Proxy can route to one or many.

Ollama logo
Ollama
Run models locally.
LM Studio logo
LM Studio
Local inference made simple.
vLLM logo
vLLM
High-throughput engine.
SGLang
Memory-efficient serving.
NVIDIA logo
NVIDIA Build
NVIDIA-hosted inference.
OpenRouter logo
OpenRouter
Access hundreds of models.
DeepSeek logo
DeepSeek
Powerful code models.
Any API
Any OpenAI-compatible API.
BUILT FOR RELIABILITY

Enterprise-grade Pool Mode

Pool Mode adds intelligent routing with failover, multi-key rotation, cooldowns, and curated logical profiles. Ship resilient AI workflows without managing complexity.

Failover

Retry another provider when one fails.

Multi-key rotation

Spread traffic across multiple API keys.

Cooldowns

Pause unhealthy providers automatically.

Logical profiles

Expose fast, balanced, and strong models.

Available Profiles
codex-fast
Low latency. Optimized for speed and completions.
codex-balanced
Best overall quality, cost, and latency. Default.
codex-strong
Max quality for complex architectural tasks.
Live Routing VisualizationACTIVE PATH
Request
Target: codex-balanced
↓
1. NVIDIA Build (Primary)
Weight: 3 • Healthy
↓
2. OpenRouter (Fallback)
Weight: 1 • Cooldown (45s)
pool.yaml
pools:
  codex-fast:
    strategy: priority
    cooldown: 60
    retries: 2
    providers:
      - openrouter: { api_key: ${KEY}, priority: 1 }
      - deepseek:   { api_key: ${KEY}, priority: 2 }

  codex-balanced:
    strategy: weighted
    providers:
      - nvidia:     { api_key: ${KEY}, weight: 3 }
      - openrouter: { api_key: ${KEY}, weight: 1 }

  codex-strong:
    strategy: priority
    providers:
      - nvidia:     { best_effort: true, priority: 1 }
      - openrouter: { api_key: ${KEY}, priority: 2 }

Why Pool Mode?

Higher uptime (failover)
Smarter cost routing
Multi-key rotation
Curated profiles

Quick Start

Get up and running in less than 60 seconds.

1. Clone & Enter
$git clone https://github.com/kyoo-147/codex_proxy_for_all_models.git
cd codex_proxy_for_all_models
2. Configure
$export CODEX_PROXY_UPSTREAM_BASE_URL=http://127.0.0.1:11434/v1
export CODEX_PROXY_UPSTREAM_MODEL=llama3.1:70b
3. Run Proxy
$python -m codex_proxy_for_all_models
~/.codex/config.toml
model = "llama3.1:70b"
model_provider = "local_model_proxy"

[model_providers.local_model_proxy]
name = "Local Model Proxy"
base_url = "http://127.0.0.1:8787"
wire_api = "responses"
env_key = "DUMMY_API_KEY"
No FastAPI. No Flask. No Node. No database. No container required.
Pure Python
Single File
Tiny Footprint
Any Env

Build the future, your way.

One proxy. Any model. Infinite possibilities.

Get Started Now →Star on GitHub