OpenCode Go Setup: Connect the Provider and Select a Model

Complete the OpenCode Go connection flow#

OpenCode’s current Go documentation describes Go as a provider that works inside the normal OpenCode workflow. Use this order:

1. Sign in to OpenCode and subscribe to Go if the account needs the subscription.
2. Copy the API key from the OpenCode account or console page.
3. Start OpenCode in the project where you want to use it.
4. Run /connect in the TUI.
5. Select OpenCode Go, then paste the key when prompted.
6. Run /models and choose a model that is listed for the connected account.

The key should be entered through the provider flow, not committed to opencode.json, a shell script, or a public issue. If the terminal keeps a command history, avoid putting the full key in a command-line argument.

Verify the provider before changing configuration#

Use the TUI to check the boundary in small steps:

1
/connect
2
/models

If /models shows the Go catalog, authentication and provider discovery are working. Select a model from that output and run a small, read-only request before assigning it to a long-running workflow.

OpenCode’s provider documentation uses a provider/model format for explicit model configuration. The Go guide currently shows IDs such as opencode-go/kimi-k3; the catalog can change, so copy the exact ID shown by your installed OpenCode version instead of keeping a model name from an old screenshot.

An explicit configuration can look like this:

1
{
2
  "$schema": "https://opencode.ai/config.json",
3
  "model": "opencode-go/kimi-k3"
4
}

Treat the model value as a checked input. If the model disappears from /models, replace the value with a currently listed ID and verify the request again. Do not add a custom provider block merely because a catalog model changed.

Keep Go limits separate from setup failures#

The Go documentation currently describes dollar-based limits for a five-hour window, a week, and a month. Those limits are usage policy, not evidence that the key or model configuration is invalid. The same page also says that the model list and limits may change as the service evolves.

Use this diagnosis order when a request fails:

If the problem is local command execution rather than provider access, keep those concerns separate. For example, Codex’s --approve-for-me boundary controls approval review and sandbox policy; connecting a model provider does not change what a local coding tool may write.

Use the current catalog as the source of truth#

OpenCode Go’s model list is intentionally not a permanent contract. Before putting a model ID in a team config:

1. Run /models with the same account that will run the workflow.
2. Record the exact opencode-go/<model-id> value.
3. Test one small request with the repository’s normal permissions.
4. Keep a fallback model or documented change procedure if the workflow is operationally important.

This makes a model change visible as a provider-catalog change instead of an unexplained application failure. It also prevents a copied model table from turning into stale configuration.

Takeaway#

The reliable OpenCode Go setup is /connect followed by /models, with the current catalog deciding the model ID. Keep the API key out of files, treat usage limits as a separate account boundary, and only pin a model after the connected account has listed and answered with it.

FAQ#