OpenCode

OpenCode can call CompactifAI through the same OpenAI-compatible chat completions API you use elsewhere: register a custom provider with our base URL and API key, then map each model id and display name for the models your account can use.

You need a CompactifAI API key and the correct model id for each model (for example carina-60b or glm-5-2). Use CompactifAI API https://api.compactif.ai/v1 unless you are explicitly targeting another environment.

1. Open the model picker

Click the model button in the bottom toolbar (it shows the currently selected model, such as a default OpenCode Zen model).

2. Manage models

In the model menu, open Manage models (settings / sliders icon next to the search field).

3. Connect provider

In the Manage models window, click + Connect provider.

4. Choose Custom

In Connect provider, open the Custom entry (under Other) to add an OpenAI-compatible endpoint.

5. Configure the custom provider

In the Custom provider form, fill in:

Field	Value
Provider ID	cai (lowercase letters, numbers, hyphens, or underscores)
Display name	CompactifAI
Base URL	https://api.compactif.ai/v1
API key	Your CompactifAI API key

Under Models, add each model your workspace should expose—for example:

• ID: carina-60b (or another catalog ID such as glm-5-2)
• Name: Carina 60B

Use + Add model for additional CompactifAI models; ID must match the API model identifier from the models catalog.

6. Submit

Click Submit to save the provider.

7. Restart OpenCode

Fully quit OpenCode and start it again. Then open the model control so you can pick from providers as before.

8. Select CompactifAI models

Scroll in the model list until you see the CompactifAI section with the models you configured.

9. Pick a model and chat

Choose a CompactifAI model from the list and start your session. Requests use the same /v1/chat/completions flow described in Chat Completion.

Image input (vision models)

OpenCode gates attachments per-model

OpenCode tracks per-model capabilities. A custom provider model created via the Custom provider form or opencode.jsonc defaults to attachment: false, so OpenCode rejects image attachments client-side with the error “this model does not support image input” — before the request reaches the CompactifAI API. This is independent of whether the model itself supports vision on the API.

To enable image input for a vision-capable model (for example qwen-3-8-27b), add attachment: true and modalities.input to the model entry in your opencode.jsonc:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "cai": {
      "name": "CompactifAI",
      "npm": "@ai-sdk/openai-compatible",
      "options": {
        "baseURL": "https://api.compactif.ai/v1"
      },
      "models": {
        "qwen-3-8-27b": {
          "name": "Qwen 3.8 27B",
          "attachment": true,
          "modalities": { "input": ["text", "image"] }
        }
      }
    }
  }
}

Restart OpenCode after editing the config. For the list of models that accept image_url content parts, see Multi-Modality.

Happy coding.

For request and response fields, see Chat Completion and the API reference.