> ## Documentation Index
> Fetch the complete documentation index at: https://docs.compressi.us/llms.txt
> Use this file to discover all available pages before exploring further.

# Connect OpenCode to CMX for Automatic Context Compression

> Step-by-step guide to routing OpenCode through the CMX local gateway so your AI coding sessions benefit from automatic context compression.

Connect OpenCode to CMX so requests from OpenCode pass through the local gateway. CMX compresses eligible context in the background, reducing your provider token usage across longer coding sessions — without requiring any changes to how you use OpenCode.

<Note>
  **Prerequisites:** CMX must be installed and signed in on your device. OpenCode must already be installed with a configured model provider. CMX does not replace your provider account; you continue managing your provider sign-in through OpenCode as usual.
</Note>

<Steps>
  <Step title="Set Up CMX">
    Run the guided setup command to sign in, start the local gateway, and detect your coding agents:

    ```bash theme={null}
    cmx setup
    ```

    Follow the browser sign-in prompt. Once setup completes, the CMX gateway starts at `127.0.0.1:17322` and is ready to accept connections.
  </Step>

  <Step title="Connect OpenCode">
    Open the CMX terminal UI and connect OpenCode to your provider:

    ```bash theme={null}
    cmx
    ```

    1. Navigate to **CONFIG → OpenCode**.
    2. **Refresh the provider list** to make sure your configured provider appears.
    3. Select the provider you use in OpenCode.
    4. Click **Connect**.
    5. Restart OpenCode so it picks up the new routing through CMX.

    OpenCode will now send all requests through the CMX gateway instead of directly to your provider.
  </Step>

  <Step title="Verify the Connection">
    Confirm that OpenCode is routing through CMX correctly:

    ```bash theme={null}
    cmx harness verify
    ```

    Send a request from OpenCode as you normally would. Then check CMX for activity:

    ```bash theme={null}
    cmx stats
    ```

    New requests should appear in the stats output. If you see activity, the connection is working.
  </Step>
</Steps>

<Note>
  Short conversations may show little or no compression. Savings grow with longer sessions as eligible older context accumulates. Compare stats across a full coding session to get a representative picture.
</Note>

## Troubleshooting

If requests from OpenCode are not appearing in `cmx stats` after a coding session:

1. **Restart OpenCode** — the new routing only takes effect after a full restart.
2. **Check the selected connection** — open CONFIG → OpenCode in CMX and confirm the correct provider is connected.
3. **Run the diagnostics command:**

   ```bash theme={null}
   cmx doctor
   ```

For more detailed guidance, see the [troubleshooting guide](/troubleshooting). If the issue persists, contact support with your CMX version, OS, and a description of what happened.

Using a tool that is not Codex or OpenCode? See [Connect Any OpenAI or Anthropic Agent to CMX](/providers/custom-agents) for instructions on connecting other tools via a custom base URL.


## Related topics

- [Connect OpenAI Codex to CMX for Context Compression](/providers/codex.md)
- [CMX Quick Start: Install, Connect, and Save Tokens](/quickstart.md)
- [Compressius Maximus (CMX): What It Is and How It Works](/introduction.md)