OpenCodex v0.1.0 NVIDIA NIM Preview

OpenCodex v0.1.0 NVIDIA NIM Preview

This is the first OpenCodex preview release focused on community-driven model provider support. The headline feature is native NVIDIA NIM integration, allowing Codex to run against NIM-hosted OpenAI-compatible chat models.

Highlights

• Added built-in nvidia-nim provider.
• Added NVIDIA NIM model discovery through /v1/models.
• Added chat inference through /v1/chat/completions.
• Added streaming assistant output for NIM chat models.
• Added Thinking display for models that stream delta.reasoning_content.
• Added support for hosted NVIDIA NIM and custom/private NIM endpoints.
• Added local build workflow so OpenCodex can run without touching global Codex state.

Usage

$env:CODEX_HOME="D:\path\to\opencodex\.codex-local"
$env:NVIDIA_API_KEY="nvapi-your-key"

& ".\codex.exe" `
  -c 'model_provider="nvidia-nim"' `
  -c 'model="z-ai/glm-5.1"'

For private NIM endpoints:

$env:CODEX_NVIDIA_NIM_BASE_URL="https://your-nim-host.example.com/v1"

Included Assets
codex-open-nvidia-nim-preview-windows-x64.zip
SHA256SUMS.txt

Known Limits

• Provider-specific controls like custom thinking budgets are not exposed yet.
• Model availability depends on your NVIDIA account, region, and endpoint.
• Image generation, web search, and OpenAI-hosted provider features are disabled for nvidia-nim.
• Tool behavior depends on each NIM model’s OpenAI-compatible tool-call support.

Community

• OpenCodex is community-shaped. If a provider, model, or workflow is missing, open an issue with official docs, model ids, auth details, and an example request. Useful requests will guide the next provider integrations.