Providers & Models

Defining a provider

[opencode.providers.nvidia]
npm = "@ai-sdk/openai-compatible"
name = "NVIDIA NIM"
base_url = "https://integrate.api.nvidia.com/v1"
api_key_env = "NIM_API_KEY"

Field	Type	Description
`npm`	string	NPM package for the provider SDK (default: `@ai-sdk/openai-compatible`)
`name`	string	Display name
`base_url`	string	API base URL
`api_key_env`	string	Environment variable holding the API key (never stored in TOML)

The api_key_env is rendered as {{env:VAR_NAME}} in the generated JSON — opencode resolves it at runtime.

Defining models

Models reference their provider by key:

[opencode.models.qwen3-5-122b]
provider = "nvidia"
model_id = "qwen/qwen3.5-122b-a10b"
name = "Qwen 3.5 122B MoE (Daily Driver)"
context_length = 262144
max_output = 16384
temperature = 0.6
top_p = 0.95
thinking = true

Field	Type	Description
`provider`	string	Provider key (must match a `[opencode.providers.X]` key)
`model_id`	string	API model identifier
`name`	string	Display name
`context_length`	integer	Context window size
`max_output`	integer	Maximum output tokens
`temperature`	float	Sampling temperature
`top_p`	float	Nucleus sampling
`top_k`	integer	Top-K sampling
`thinking`	boolean	Enable chain-of-thought / thinking mode

Setting the default model

[opencode.default_model]
model = "qwen3-5-122b"

This resolves to nvidia/qwen/qwen3.5-122b-a10b in the generated JSON.

Generated output

Models are nested inside their provider in the JSON:

{
  "provider": {
    "nvidia": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "NVIDIA NIM",
      "options": { "baseURL": "..." },
      "models": {
        "qwen/qwen3.5-122b-a10b": {
          "name": "Qwen 3.5 122B MoE (Daily Driver)",
          "limit": { "context": 262144, "output": 16384 },
          "parameters": { "temperature": 0.6, ... }
        }
      }
    }
  }
}

Overview

MCP Servers