OpenCode

Status: Candidate setup, not yet verified by NexoRouter.

OpenCode supports custom OpenAI-compatible providers. Configure NexoRouter as a provider that calls https://api.nexorouter.com/v1 and sends Chat Completions requests.

What this setup uses

Do not use this page to configure Codex CLI, Claude Code, or Gemini-native clients. Those tools have separate protocol requirements.

Before you start

1. Install OpenCode from the current OpenCode install docs.
2. Run opencode --version to confirm it is available.
3. Create a NexoRouter API key with a low budget for the first test.
4. Copy one model ID from Models, for example deepseek-v4-flash.
5. Keep the key out of opencode.json, Git, screenshots, and support messages.

OpenCode references:

• OpenCode intro
• OpenCode providers
• OpenCode config

Credential flow

OpenCode's provider docs use the /connect command to add provider credentials. If your version supports it:

/connect

Then choose Other, enter a provider ID such as nexorouter, and paste your NexoRouter API key.

If your OpenCode version uses a different credential flow, follow that version's provider docs. The important part is that the secret is stored in OpenCode's credential storage, not committed in project config.

Provider configuration

Use a stable provider ID such as nexorouter.

Global config is commonly stored at:

~/.config/opencode/opencode.json

Example:

{
  "provider": {
    "nexorouter": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "NexoRouter",
      "options": {
        "baseURL": "https://api.nexorouter.com/v1"
      },
      "models": {
        "deepseek-v4-flash": {
          "name": "DeepSeek V4 Flash"
        },
        "gpt-4o-mini": {
          "name": "GPT-4o mini"
        },
        "Qwen/Qwen-Plus": {
          "name": "Qwen Plus"
        }
      }
    }
  },
  "model": "nexorouter/deepseek-v4-flash"
}

If your OpenCode version requires the key in config, prefer an environment variable reference if supported by that version. Do not hard-code the full key in a repository file.

First test

Start with a read-only task:

Read the current directory and summarize the project in one paragraph. Do not edit files.

Then open Usage Logs and confirm:

• API key name.
• Model ID.
• Request status.
• Prompt and completion tokens.
• Cost and latency.
• Request ID.

Second test

Only after the read-only task appears in Usage Logs, test one small edit in a disposable branch:

Create a short README note in a temporary branch. Do not change existing source files.

If the text prompt works but the edit workflow fails, treat that model/tool pair as not ready for real coding-agent work.

Common issues

Known limits

• Treat this page as a candidate setup until your current OpenCode version passes the read-only test, edit test, and Usage Logs check.
• Streaming and tool calling behavior can vary by model and OpenCode version.
• Use a low-budget key for local agents.
• Do not reuse a personal production key in shared repositories or demos.

Related pages

• Tool calling
• Usage Logs
• API key invalid