Using Aegis With Agents

Configure OpenCode, Cline, and Claude Code Router with one Aegis key, and understand the current Codex limitation.

OpenCode

OpenCode talks to Aegis through the `@ai-sdk/openai-compatible` adapter, registered as a custom provider. The setup is four short steps: install OpenCode, generate an Aegis key, write the config file, then launch and pick a model.

Each step below has the exact command or file you need. Follow them in order.

Never commit your `aegis_...` key to version control. Keep it in `opencode.jsonc` only, or rotate the key immediately if it leaks.

Step 1: Install

Install OpenCode from the official installer. The command below works on macOS, Linux, and WSL.

Run the following command:

  • Verify the install with `opencode --version`.
  • Prefer a modern terminal (WezTerm, Alacritty, Ghostty, Kitty) for the best experience.
curl -fsSL https://opencode.ai/install | bash

Step 2: Generate An Aegis API Key

You need an Aegis API key before configuring OpenCode. The full `aegis_...` value is shown only once at creation time.

  • Sign in to Aegis.
  • Open Console > API Keys.
  • Enter a label such as `opencode` or `Local dev`.
  • Click `Generate Key`, then copy the `aegis_...` value immediately.
  • Store the key somewhere safe — you will not be able to see it again.

Step 3: Configure The Custom Provider

Open (or create) the OpenCode config file and register Aegis as a custom provider. Replace `your_aegis_api_key_here` with the key you copied in Step 2.

Run the following command to open the config file:

  • Create the config directory if it does not exist: `mkdir -p ~/.config/opencode`.
  • Copy the following snippet and paste it into the file.
  • Replace the `apiKey` value with your real `aegis_...` key from Step 2.
  • Set `"model"` to the model ID you want as the default — for example `aegis/glm-5.2`.
  • Save and exit (`Ctrl-O`, `Enter`, `Ctrl-X` in nano).
The `apiKey` field accepts a raw string. If you prefer to avoid hardcoding it, you can also reference an environment variable using the documented OpenCode placeholder syntax.
nano ~/.config/opencode/opencode.jsonc

~/.config/opencode/opencode.jsonc

{
  "$schema": "https://opencode.ai/config.json",
  "model": "aegis/glm-5.2",
  "provider": {
    "aegis": {
      "name": "Aegis",
      "npm": "@ai-sdk/openai-compatible",
      "options": {
        "baseURL": "https://api.aegisrouter.com/v1",
        "apiKey": "your_aegis_api_key_here"
      },
      "models": {
        "glm-5.2": {
          "name": "GLM-5.2",
          "limit": {
            "context": 1048560,
            "output": 128072
          }
        }
      }
    }
  }
}

Step 4: Launch And Pick A Model

You are ready to use OpenCode against Aegis. Launch it from a project directory and switch models from the built-in model picker when needed.

Run the following commands:

  • Initialize it for the project on first run: `/init`.
  • To switch models, use the model picker (default keybinding `/`) and select `aegis/glm-5.2`.
  • Send a prompt to confirm traffic flows through Aegis.
Check the Aegis dashboard > Usage to verify requests are landing. If nothing shows up, re-check that `apiKey` in `opencode.jsonc` is your real `aegis_...` value and not the placeholder.
cd /path/to/project
opencode

# In the OpenCode TUI:
#   /init          # create AGENTS.md on first run
#   /model         # switch to aegis/glm-5.2

Cline

Cline supports OpenAI-compatible providers directly from its settings panel.

  • Open the Cline settings panel.
  • Set `API Provider` to `OpenAI Compatible`.
  • Set `Base URL` to `https://proxy.aegis.dev/v1`.
  • Paste your `aegis_...` key into the API key field.
  • Enter a model ID from the Aegis Models page, such as `glm-5.2` or `qwen3.6-35b`.
If Cline asks for optional model metadata such as context window, pricing, or image support, use the values listed on the Aegis Models page.

Claude Code

Claude Code does not speak the OpenAI format directly. Route it through `claude-code-router`, then send that router to Aegis's chat completions endpoint.

Shell

npm install -g @anthropic-ai/claude-code
npm install -g @musistudio/claude-code-router

export AEGIS_API_KEY="aegis_..."

~/.claude-code-router/config.json

{
  "Providers": [
    {
      "name": "aegis",
      "api_base_url": "https://proxy.aegis.dev/v1/chat/completions",
      "api_key": "$AEGIS_API_KEY",
      "models": ["glm-5.2", "qwen3.5-397b", "kimi-k2.7-code"]
    }
  ],
  "Router": {
    "default": "aegis,glm-5.2",
    "background": "aegis,glm-5.2"
  }
}

Codex CLI

Codex CLI currently expects the OpenAI Responses API. Aegis currently documents and supports the OpenAI-compatible chat completions shape, not a direct Responses API integration.

That means you should not point Codex CLI at Aegis yet and expect full support. Use OpenCode, Cline, or Claude Code Router today for coding-agent workflows over the Aegis endpoint.

When Aegis adds a supported Responses API surface, this page should be expanded with a `~/.codex/config.toml` example.