Skip to content

Integration guide

Use GreenPT in Codex

OpenAI Codex CLI supports custom model providers, so it can run on GreenPT. Add a greenpt provider to ~/.codex/config.toml with the GreenPT base URL and your API key, set glm-5.2 as the model, and Codex plans, edits, and executes on renewable EU infrastructure.

Codex reads the provider from one TOML file; no forks, wrappers, or patched binaries involved.

Setup

Connect Codex CLI to GreenPT

Every OpenAI-compatible setup needs the same three values. The steps below show where they go in this tool.

Base URL
https://api.greenpt.ai/v1
API key
GREENPT_API_KEY
Model
glm-5.2
  1. Install Codex CLI

    Install the CLI globally with npm (or Homebrew: brew install codex). Any recent version supports custom providers.

    Terminal
    npm install -g @openai/codex
  2. Create a GreenPT API key

    Sign up at account.greenpt.ai (14-day free trial, no credit card) and create an API key in the console.

  3. Add GreenPT as a model provider

    Declare the provider and default model in ~/.codex/config.toml. The wire_api = "chat" line tells Codex to use the OpenAI-compatible chat completions endpoint.

    ~/.codex/config.toml
    model = "glm-5.2"
    model_provider = "greenpt"
    
    [model_providers.greenpt]
    name = "GreenPT"
    base_url = "https://api.greenpt.ai/v1"
    env_key = "GREENPT_API_KEY"
    wire_api = "chat"
  4. Export your key and run Codex

    Codex reads the key from the env_key variable named in the config. Start it in a project and give it a small task to verify the model responds.

    Terminal
    export GREENPT_API_KEY=your_key_here
    codex

Pick a model

Five open coding models, one key

glm-5.2 is the flagship for agentic software engineering. kimi-k2.7-code handles very large repositories, devstral-2-123b-instruct-2512 is built for SWE agents, qwen3-coder-30b-a3b-instruct is the efficiency pick, and minimax-m2.5 the budget one. All open-weight, all hosted in the EU.

Compare all coding models

Why GreenPT

Why teams point Codex CLI at GreenPT

Same agent, same workflow, different footprint. GreenPT serves open-weight coding models from green EU data centres, with privacy guarantees that proprietary defaults do not offer.

  • No training on your code

    Prompts and code are never used for training. GDPR-compliant, EU-based processing with DPAs on request.

  • Open-weight models

    glm-5.2, Kimi K2.7 Code, Devstral 2, Qwen3 Coder, and MiniMax M2.5. You know exactly what you run.

  • Renewable EU infrastructure

    Scaleway green data centres, 100% renewable energy, PUE 1.25 and WUE 0.25, with CO2 tracked per request.

  • One key for everything

    The same GREENPT_API_KEY also serves chat, embeddings, reranking, OCR, speech-to-text, and web search.

Frequently asked questions

Which GreenPT model should I use in Codex?

glm-5.2 is the recommended default: agentic planning, reliable function calling, and a 1M-token context. qwen3-coder-30b-a3b-instruct is the efficient pick when you want the lowest energy and cost per request.

Compare the coding models →
Does this replace my OpenAI account?

Only for model calls. The config above routes Codex requests to GreenPT; remove the model_provider line to switch back. You can also keep both and select the provider per profile.

Does GreenPT train on my code?

No. GreenPT never trains models on your prompts, code, or conversations. Processing happens entirely inside the EU with full GDPR compliance, and data processing agreements are available on request.

How is this more sustainable than the default setup?

GreenPT models run in Scaleway green EU data centres on 100% renewable energy, with a PUE of 1.25 and a WUE of 0.25, well below industry averages. CO2 and energy usage are tracked per request, so you can report the footprint of your coding agent instead of guessing it.

What does it cost?

You pay per token through GreenPT API credits, and every new account starts with a 14-day free trial, no credit card required. Open-weight models are markedly cheaper to run than proprietary frontier models, and you can cut spend further with the Honey agent skill.

View pricing →

Start coding greener

Your agent, our models, less carbon .

Start a free 14-day trial, no credit card. Point Codex CLI at one OpenAI-compatible endpoint and run glm-5.2 and friends on 100% renewable energy.

  • 100% Renewable
  • EU Hosted
  • OpenAI-compatible