Opencode Desktop configuration

Step-by-step instructions to configure OpenCode Desktop on to use the ElettraGateway LLM gateway (https://llm-api-gw.elettra.eu/v1) and its models.

All steps below are for your Windows user account. <you> means your existing user name used to log-in to your machine. The username determine the paths, e.g. C:\Users\name.surname\...on Windows or /home/name.surname on Linux.

---

Step 1 — Create the configuration file opencode.jsonc

OpenCode reads its global configuration from the file:

C:\Users\<you>\.config\opencode\opencode.jsonc

Create the .config\opencode folder and the opencode.jsonc file if they do not exist yet:

Paste the following content into opencode.jsonc and save it:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "elettragateway": {
      "name": "ElettraGateway",
      "npm": "@ai-sdk/openai-compatible",
      "options": {
        "baseURL": "https://llm-api-gw.elettra.eu/v1"
      },
      "models": {
        "balanced": {
          "name": "balanced",
          "limit": {
            "context": 262144,
            "input": 262144,
            "output": 100000
          },
          "modalities": {
            "input": ["text", "image"]
          }
        },
        "top": {
          "name": "top",
          "limit": {
            "context": 1000000,
            "input": 1000000,
            "output": 100000
          },
          "modalities": {
            "input": ["text", "image"]
          }
        },
        "fast": {
          "name": "fast",
          "limit": {
            "context": 131000,
            "input": 131000,
            "output": 100000
          }
        }
      }
    }
  }
}

What this does

• Registers a custom, OpenAI-compatible provider named ElettraGateway pointing at the gateway URL.
• Declares the available models and their context limits (limit.context, limit.input, limit.output) — this is what OpenCode uses for context accounting.
• modalities.input: ["text", "image"] marks the models that accept images.

No API-KEY is stored in this file.

---

Step 2 — Start (or reload) OpenCode Desktop

Launch OpenCode Desktop.

If the app was already running while you edited opencode.jsonc, please restart it.

---

Step 3 — Enable the models

The desktop app keeps its own list of which models appear in the model picker. Models that are not enabled there are hidden, even when they are correctly configured.

1. Click on the "three rows icon" (see screenshot), select File -> Settings.
2. Go to Providers — verify that ElettraGateway appears.
3. Go to Models, Switch off everything not belonging to ElettraGateway and switch on the ElettraGateway models (top, balanced, `fast), or use Enable all models.

---

Step 4 — Set the api key

1. Go back to the main Opencode Desktop window clicking on the "back to app" on the top left corner.
2. Type /connect as in the screenshot below, select it and click enter.
3. Type ElettraGateway as in the screenshot below, select the gateway from the services by cliking on it.
4. Paste the provided apikey in the appearing dialog window and press continue.

---

Step 5 — Testing

1. You should be on the main Opencode Desktop window.
2. The main screen presents 3 elements (refer to the screenshot below):

• a: the selection box, allowing to switch between the build/plan phase; start in plan mode to analyse and plan the work, switch to build to actually edit existing files, perform actions, etc.
• b: when clicked, the model selection box allow you choose between the configured models. Top should already be the default.
• c: Opencode utilises "workspaces" to organise the work. A workspace is an existing folder in the filesystem. Opencode will edit, write new files within that folder and will not asks for permissions.

3. Set the project workspace fodler, pick one of the ElettraGateway models, e.g. top or balanced.

4. Send a test prompt "hello" and check that a reply arrives.