Skip to main content

Model Setup & NVIDIA Credentials

M31A uses NVIDIA NIM as its exclusive production model provider.

Obtaining Credentials​

  1. Visit the NVIDIA API Catalog at build.nvidia.com.
  2. Sign in and generate an API key with NIM model access.
  3. Your key will have the format: nvapi-....

Configuring Credentials​

M31A supports three ways to configure your key, evaluated in precedence order:

export NVIDIA_API_KEY="nvapi-your-key-here"

2. User Configuration (~/.config/m31a/config.toml)​

m31a config set provider.nvidia_nim.api_key "nvapi-your-key-here"

3. TUI Settings Editor​

In the interactive cockpit, press Ctrl+, to open the Settings surface, navigate to Provider Credentials, enter your key, and press Enter. The key is masked immediately (nvap...[MASKED]).

Canonical Default Model​

The default model is defined in src/config/canonical.rs:

model = "meta/llama-3.2-11b-vision-instruct"

Supported NVIDIA NIM models include:

  • meta/llama-3.2-11b-vision-instruct (Default reasoning & vision)
  • meta/llama-3.1-70b-instruct (Heavy synthesis & architecture)
  • meta/llama-3.1-405b-instruct (Large-scale decomposition)
  • nvidia/llama-3.1-nemotron-70b-instruct (High-precision alignment)

To select a different model for a run:

m31a run --model "meta/llama-3.1-70b-instruct" "Refactor the authentication module"

In the TUI, switch models dynamically using the /model slash command.