Hermes Agent + Xiaomi MiMo: BYOK Setup Guide

What Is Xiaomi MiMo?

MiMo is Xiaomi's open LLM platform (api.xiaomi.com). The headline model, MiMo V2 Pro, is a trillion-parameter flagship explicitly tuned for agent workflows — long-context tool calling, structured outputs, code execution, and multi-step planning. Current model line:

• xiaomi/mimo-v2-pro — flagship, agent-tuned
• xiaomi/mimo-v2.5 — mid-tier, fast
• xiaomi/mimo-v2.5-pro — balance of speed + reasoning
• xiaomi/mimo-v2 — legacy V2 (cheapest)

Why Use MiMo With Hermes Agent?

• Agentic tuning — MiMo V2 Pro's training corpus includes heavy tool-use + browser + code trajectories, which matches what Hermes needs.
• Cheap relative to GPT-5 / Claude 4.6 / Gemini 2.5 Pro on long agent runs.
• OpenAI-compatible API surface — works through OpenRouter or direct provider config.

Option 1: Through OpenRouter (Recommended)

MiMo is on OpenRouter under xiaomi/*. If you already have an OpenRouter key in Hermes, the only change is the model ID:

# ~/.hermes/config.yaml
models:
  primary: openrouter/xiaomi/mimo-v2-pro
  providers:
    openrouter:
      apiKey: ${OPENROUTER_API_KEY}

Restart Hermes. The agent now plans + tool-calls through MiMo V2 Pro.

Option 2: Custom OpenAI-Compatible Endpoint

MiMo's API is OpenAI-compatible, so on self-hosted Hermes you can register it as a custom provider and skip the OpenRouter middleman:

# ~/.hermes/config.yaml
models:
  primary: mimo/mimo-v2-pro
  providers:
    mimo:
      type: openai-compatible
      apiKey: ${MIMO_API_KEY}
      baseUrl: https://api.xiaomi.com/v1

Direct routing is cheaper per token but skips OpenRouter's automatic failover. Pick OpenRouter if reliability matters more than price. On managed Hermes (OpenClaw Launch), MiMo is wired through OpenRouter today — the custom-endpoint path is self-host only.

Hermes on OpenClaw Launch + MiMo

For managed Hermes hosting, deploy from the Hermes Hosting page and set BYOK in your API Keys dashboard. The container picks up the key, no config file editing.

1. Deploy Hermes from the dashboard.
2. Open /api-keys, paste your MiMo or OpenRouter key.
3. Pick xiaomi/mimo-v2-pro as the primary model.
4. Hermes restarts the gateway, ready to use.

Picking a MiMo Variant

Verifying It Works

curl -X POST http://localhost:8080/v1/chat/completions \
  -H "Authorization: Bearer $HERMES_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"model": "xiaomi/mimo-v2-pro", "messages": [{"role":"user","content":"List files in current directory"}]}'

If the response invokes a shell tool call, MiMo is wired correctly.

Limits and Gotchas

• MiMo's tool-call schema is OpenAI-compatible but stricter on JSON validation — malformed tool args may be rejected outright instead of auto-corrected.
• Region: the direct MiMo endpoint is hosted in China. From outside CN, prefer OpenRouter.
• Rate limits on the free tier are aggressive — for production Hermes runs, use a paid MiMo token plan or OpenRouter.

What's Next?

• OpenClaw + Xiaomi MiMo — same wiring for the OpenClaw side
• Hermes Agent BYOK — full BYOK options
• Hermes Agent + Qwen — the other major Chinese open-platform option
• Hermes Agent + OpenRouter — OpenRouter routing details