OpenClaw Adds Native Arcee AI Support With Three Trinity Models Out of the Box

OpenClaw has expanded its roster of first-party AI providers with the addition of Arcee AI, shipping as a bundled plugin that brings three models from the company's Trinity lineup directly into the platform. The integration went live over the weekend, giving self-hosted users a new set of reasoning and general-purpose models without any manual plugin installation.

Three Models, Three Roles

The Arcee AI plugin arrives with support for the full Trinity family:

• Trinity Large Thinking — a reasoning-first model with a 256K context window, designed for multi-step planning, code generation, and complex agentic workflows
• Trinity Large Preview — a 128K context general-purpose model suited for conversational agents and content generation
• Trinity Mini 26B — a compact, fast-inference model with 128K context, optimized for high-throughput scenarios where cost efficiency matters more than raw reasoning power

The lineup gives users a natural tiering strategy: route demanding tasks to Trinity Large Thinking, handle everyday interactions with Trinity Large Preview, and lean on Trinity Mini 26B for quick lookups and simple automations.

Dual Authentication Paths

One notable design choice in this release is the dual authentication model. Users can authenticate directly with Arcee AI using an ARCEEAI_API_KEY, or route through OpenRouter using an existing OPENROUTER_API_KEY. For teams already using OpenRouter to manage multiple providers under a single billing umbrella, the second path means Arcee models are available immediately — no new vendor account required.

This is part of a broader pattern in OpenClaw's provider architecture: environment variables are now dynamically computed from manifest-driven provider and channel declarations, replacing previous hardcoded key lists. The change means future provider plugins can ship with their own credential requirements without touching core configuration code.

What Changed Under the Hood

Beyond the plugin itself, the integration prompted a refactor of how OpenClaw handles provider manifests. Channel manifests now publish environment-driven setup keys, and the shell-env fallback system computes expected keys dynamically rather than relying on a static list. This is a subtle but important shift: it means the platform's configuration layer scales automatically as new providers are added.

The model catalog system was also updated to support manifest-level augmentation, so provider plugins can declare their models declaratively rather than requiring imperative registration.

Who Is Arcee AI?

Arcee AI has been building enterprise-focused language models with an emphasis on domain adaptation and efficient inference. The Trinity family represents their latest general-purpose offering, positioned as a competitive alternative to models from Anthropic, OpenAI, and Google — particularly for organizations that want strong reasoning performance without the price tag of frontier models.

The 256K context window on Trinity Large Thinking is particularly relevant for agentic use cases, where long conversation histories and multi-document analysis are common.

Getting Started

For existing OpenClaw users, the Arcee AI plugin is already included in the latest build. Set your ARCEEAI_API_KEY or OPENROUTER_API_KEY environment variable, and the Trinity models will appear in your model selector automatically.

New to OpenClaw? The fastest way to get started is through DeployClaw, which handles deployment, updates, and configuration — including provider credentials — so you can focus on building agents rather than managing infrastructure.