OpenCode

OpenCode and OpenCode Zen expose models through several API formats. Choose the format that matches the model or gateway, then use the resulting provider like any other nimgent provider.

Create a model

This example uses the default Chat Completions format.

import std/os
import nimgent
import nimgent/providers/openai

let model = openCode(
  getEnv("OPENCODE_API_KEY"),
  protocol = ocChat
).model(getEnv("OPENCODE_MODEL"))

echo generateText(model, prompt = "Say hello in one sentence.").text

Set OPENCODE_API_KEY and OPENCODE_MODEL before you run the program. The repository does not define an OpenCode model ID, so use one from the OpenCode catalog.

Choose the request format

Set protocol to match the model’s gateway format:

Protocol	Use when the gateway expects
ocChat	OpenAI Chat Completions. This is the default.
ocResponses	OpenAI Responses.
ocMessages	Anthropic Messages.
ocGoogle	Native Google Gemini paths.

For OpenCode Zen, use openCodeZen with the same protocol choices:

let model = openCodeZen(
  getEnv("OPENCODE_API_KEY"),
  protocol = ocResponses
).model(getEnv("OPENCODE_MODEL"))

If a request fails, confirm that the selected protocol matches the model’s gateway. ocGoogle is for Gemini models served on native Google API paths.

Next steps

See Settings for portable controls, or Middleware and routing to choose providers from a model ID.