OpenCode

opencode.json in the project root, or ~/.config/opencode/opencode.json for every project. OpenCode merges the two; the project file wins on keys that clash.

Keys saved with /connect live in ~/.local/share/opencode/auth.json.

OpenCode only takes an OpenAI-compatible base URL for a custom provider, so use the @ai-sdk/openai-compatible package. Every id under models shows up in the /models picker. Any id on the Models page works, Claude included.

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "uplinkswitch": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "UplinkSwitch",
      "options": {
        "baseURL": "https://api.uplinkswitch.com/v1",
        "apiKey": "{env:UPLINKSWITCH_API_KEY}"
      },
      "models": {
        "claude-sonnet-5-5": { "name": "Claude Sonnet 5.5" },
        "claude-opus-5-5": { "name": "Claude Opus 5.5" },
        "gpt-6.1-sol": { "name": "GPT-6.1 Sol" },
        "gemini-3.1-pro-preview": { "name": "Gemini 3.1 Pro" },
        "deepseek-v4-pro-0813": { "name": "DeepSeek V4 Pro" },
        "kimi-k3": { "name": "Kimi K3" }
      }
    }
  },
  "model": "uplinkswitch/claude-sonnet-5-5",
  "small_model": "uplinkswitch/claude-haiku-5-5"
}

model and small_model take provider/model. Each model entry can also carry limit.context and limit.output (tokens in, tokens out) if you want OpenCode to budget context for it.

{env:UPLINKSWITCH_API_KEY} reads the variable at start. An unset variable becomes an empty string, so export it before you run OpenCode:

export UPLINKSWITCH_API_KEY="sk-uplink-live-..."
opencode

Or skip options.apiKey and store the key with /connect: scroll to Other, enter the provider id uplinkswitch, paste the key. The id must match the key you used under provider.

Confirm the key before you open the TUI:

curl https://api.uplinkswitch.com/v1/models -H "Authorization: Bearer $UPLINKSWITCH_API_KEY"

Then run opencode and /models. UplinkSwitch and the ids you listed appear in the picker. opencode auth list shows the stored credential if you used /connect.

• @ai-sdk/openai-compatible calls /v1/chat/completions. OpenCode reserves @ai-sdk/openai for /v1/responses; use the compatible package here.
• Only the ids you list are offered. Add a new one to models when the Models page gains it.
• small_model handles light work such as titles. Leave it out and OpenCode picks a cheaper model from the provider if it finds one, otherwise the main model.
• blacklist and whitelist on the provider hide or keep ids in /models, by the same ids.