Using Arcee AI with OpenClaw

Prerequisites

• ✓An Arcee AI account with an API key from https://chat.arcee.ai/ or an OpenRouter account with an API key from https://openrouter.ai/keys.
• ✓An existing OpenClaw installation with access to the `openclaw` CLI.
• ✓Access to the machine or environment where the OpenClaw Gateway runs so you can persist environment variables (for example in `~/.openclaw/.env`).

Steps

1. 1

   Decide whether to use Arcee directly or via OpenRouter

   Arcee AI models work either through Arcee’s own API or via OpenRouter, and the rest of your setup depends on this choice. Direct access uses `ARCEEAI_API_KEY` and the Arcee base URL, while OpenRouter uses `OPENROUTER_API_KEY` and the OpenRouter base URL.

   Pick the path that matches how you already consume Arcee or other models.

   textCopy

   Arcee AI models can be accessed directly via the Arcee platform or through [OpenRouter](https://docs.openclaw.ai/providers/openrouter).
   *   Provider: `arcee`
   *   Auth: `ARCEEAI_API_KEY` (direct) or `OPENROUTER_API_KEY` (via OpenRouter)
   *   API: OpenAI-compatible
   *   Base URL: `https://api.arcee.ai/api/v1` (direct) or `https://openrouter.ai/api/v1` (OpenRouter)

2. 2

   Get your Arcee or OpenRouter API key

   You need a valid API key before OpenClaw can talk to Arcee’s Trinity models. Grab it from Arcee if you’re going direct, or from OpenRouter if you want to route through their gateway.

   Keep it handy in your shell environment so you can pass it into the onboarding command or store it in the Gateway.

   textCopy

   1.   Get an API key from [Arcee AI](https://chat.arcee.ai/) or [OpenRouter](https://openrouter.ai/keys).

3. 3

   Run interactive onboarding for Arcee AI

   Use the interactive `openclaw onboard` flow when you want OpenClaw to prompt you and store credentials for the Gateway. There are separate presets for direct Arcee and for OpenRouter, and picking the wrong one is a common source of auth errors.

   Run the command that matches your earlier choice so OpenClaw wires up the right auth and base URL.

   bashCopy

   # Direct (Arcee platform)
   openclaw onboard --auth-choice arceeai-api-key
   
   # Via OpenRouter
   openclaw onboard --auth-choice arceeai-openrouter

4. 4

   Run non-interactive onboarding in automated environments

   For CI, scripts, or headless servers, use the non-interactive onboarding variant so everything can run unattended. You pass the mode, auth choice, and the API key via flags, which avoids prompts and makes the setup reproducible.

   Make sure the corresponding environment variable is set in your shell before running these commands.

   bashCopy

   # Direct (Arcee platform)
   openclaw onboard --non-interactive \
     --mode local \
     --auth-choice arceeai-api-key \
     --arceeai-api-key "$ARCEEAI_API_KEY"
   
   # Via OpenRouter
   openclaw onboard --non-interactive \
     --mode local \
     --auth-choice arceeai-openrouter \
     --openrouter-api-key "$OPENROUTER_API_KEY"

5. 5

   Set a Trinity model as the default for your agents

   Once auth is wired up, configure a default model so your agents automatically use Arcee’s Trinity models. The onboarding preset already uses `arcee/trinity-large-thinking`, but it helps to know the exact config shape if you want to edit it by hand.

   This config makes Trinity Large Thinking the primary model for all default agents.

   jsonCopy

   {
     agents: {
       defaults: {
         model: { primary: "arcee/trinity-large-thinking" },
       },
     },
   }

6. 6

   Choose the right Trinity model from the built-in catalog

   OpenClaw ships a built-in catalog of Arcee models so you can switch between them by reference. Each model has different context limits and cost profiles, and the same refs work whether you connect directly or via OpenRouter.

   Use these refs in your agent configs when you need a faster, cheaper, or more reasoning-focused model.

   textCopy

   | Model ref | Name | Input | Context | Cost (in/out per 1M) | Notes |
   | --- | --- | --- | --- | --- | --- |
   | `arcee/trinity-large-thinking` | Trinity Large Thinking | text | 256K | 0.25/0.25 / 0.25/0.90 | Default model; reasoning enabled |
   | `arcee/trinity-large-preview` | Trinity Large Preview | text | 128K | 0.25/0.25 / 0.25/1.00 | General-purpose; 400B params, 13B active |
   | `arcee/trinity-mini` | Trinity Mini 26B | text | 128K | 0.045/0.045 / 0.045/0.15 | Fast and cost-efficient; function calling |

7. 7

   Ensure API keys are available to the Gateway process

   If your OpenClaw Gateway runs under launchd or systemd, environment variables from your login shell often don’t propagate. You need to make sure `ARCEEAI_API_KEY` or `OPENROUTER_API_KEY` is visible to the daemon, otherwise calls to Arcee will fail at runtime.

   env` or configure them via your process manager’s environment settings.

   textCopy

   If the Gateway runs as a daemon (launchd/systemd), make sure `ARCEEAI_API_KEY` (or `OPENROUTER_API_KEY`) is available to that process (for example, in `~/.openclaw/.env` or via `env.shellEnv`).

Configuration

Troubleshooting

{
  agents: {
    defaults: {
      model: { primary: "arcee/trinity-large-thinking" },
    },
  },
}

Frequently asked questions