# Use OpenCode Go with Agent Zero OpenCode Go is available through a community provider plugin. If you cannot find it in the Cloud provider grid, open the Plugin Hub from that same screen, install **OpenCode Go**, and choose its models in **Advanced model configuration**. This guide starts with a fresh Agent Zero instance. The same route helps you find other providers supplied as plugins. ## Before you start You can install the plugin and explore the settings without a subscription. To send messages through Go, you need your own OpenCode API key and an active [OpenCode Go subscription](https://opencode.ai/docs/go/). You do not need to install the OpenCode app to use this provider in Agent Zero. The screenshots below stop at model configuration, with the key fields empty. Installation and model selection were checked with OpenCode Go plugin 0.1.4; model responses were not tested. ## 1. Send your first message Open Agent Zero and type a message in the welcome screen's composer. For example: **Help me plan a weekend hike.** Click the arrow to send it. ![Welcome composer with a first message ready to send](../res/usage/opencode/welcome-message.png) Agent Zero opens a chat and holds your message while you choose a model. Click **Cloud provider**. ![The chat setup gate with Cloud, account, local and provider-plugin choices](../res/usage/opencode/model-gate.png) ## 2. Find the provider-plugin link OpenCode Go is not in the Cloud grid. Scroll to the bottom and click the puzzle link: **Didn't find your provider? Click to look in the Plugin Hub**. ![The plugin link underneath the last rows of Cloud providers](../res/usage/opencode/cloud-plugin-link.png) The **Account** and **Local** tabs have the same link. You can also go straight to the Hub using **or click here to browse provider plugins** below the three choices in the chat. ## 3. Install OpenCode Go The Hub opens with **Provider** selected. Search for **OpenCode**, then open **OpenCode Go** by **3clyp50**. ![OpenCode Go found in the Plugin Hub with the Provider filter selected](../res/usage/opencode/provider-hub.png) Read the description and README, and check the [plugin's source](https://github.com/3clyp50/a0_opencode_go). It adds both Go and Zen providers. This walkthrough uses Go. ![OpenCode Go plugin details with the Install button](../res/usage/opencode/install-plugin.png) Click **Install**. Read the third-party plugin warning. If you trust the plugin, choose **Install Anyway**. Wait until the page shows **Installed**, then refresh Agent Zero in your browser. If the plugin already says **Installed**, skip installation and refresh the page. You do not need to reinstall it. ## 4. Choose the Main model Return to your chat. The held message and setup choices remain after the refresh. Click **Advanced model configuration** at the bottom of the setup message. This opens **Edit Model Presets**. Leave **Preset** set to **Default** for this first setup. In **Main Model**: 1. Set **Provider** to **OpenCode Go (GLM/Kimi/DeepSeek/MiMo)**. Choose it by name; the other OpenCode entries serve different model groups or Zen. 2. Replace the old **Model name** with `glm-5.3`. You can type it or use the search icon and select it from the list. Changing the provider alone does not clear the previous model name. 3. When you are ready to connect, paste your OpenCode key into **API key**. ![Default preset with OpenCode Go and glm-5.3 selected for Main Model](../res/usage/opencode/main-model.png) This example is for text chats. Turn **Supports Vision** off and leave the separate vision settings empty for now. You can configure image analysis later in [Model Presets](model-presets.md). ## 5. Choose the Utility model Scroll to **Utility Model**. This model handles short background tasks such as summaries and chat names. Choose the same Go provider, then replace the old model name with `glm-5.3-flash`. ![Utility Model using OpenCode Go and glm-5.3-flash with an empty API key field](../res/usage/opencode/utility-model.png) | Section | Model name | | --- | --- | | Main Model | `glm-5.3` | | Utility Model | `glm-5.3-flash` | These are the IDs shown by Agent Zero's model picker and [OpenCode's Go model list](https://opencode.ai/docs/go/#endpoints). Enter them exactly, without a `zai/`, `open/`, or `opencode-go/` prefix. The free `glm-5.3-flash-free` model is a different offering. OpenCode's [Console documentation](https://opencode.ai/docs/zen/#endpoints) says its free chat models require the OpenCode client. Use the Go model above for this setup; it is covered by your Go access, not a free Utility model. Both sections share the API key when they use the same provider. Leave the fresh instance's **Embedding Model** set to **HuggingFace** and `sentence-transformers/all-MiniLM-L6-v2`. ## 6. Save when you are ready to connect Check the provider, model names, and your key, then click **Save**. Return to the chat. Once the model is configured, Agent Zero can retry your held message automatically. Do not send it again while it is starting. If you are only exploring and do not have Go access yet, click **Close** instead. The plugin stays installed, but the unfinished preset changes are not saved. Later, reopen **Advanced model configuration** from the chat. You can also open the preset menu above the composer and choose **Edit Presets**. ## If something is missing - **No OpenCode provider in the editor:** check that OpenCode Go is installed and enabled under **Plugins → Custom**, then refresh the browser page. - **A model is missing from the search results:** check OpenCode's current model list and the plugin's updates. The Model name field also accepts an exact model ID typed by hand. Typing a name does not grant access to it. - **A key or access error:** check the key and your Go subscription in the [OpenCode console](https://opencode.ai/console). Installing the plugin does not create an account or subscription. - **Looking for another provider:** return to **Plugins → Browse**, select **Provider**, and search for its name. Follow that plugin's README for its provider choice, model names, and key requirements. For the other setup routes, see [First-Run Onboarding](onboarding.md).