Moonshot AI and Dome

1. 01

   Add the connections

   The provider id is moonshot. Dome uses https://api.moonshot.ai/v1 and Moonshot's own model ids; Fireworks uses its own.

   $ dome models add kimi-moonshot \
     --provider moonshot \
     --model kimi-k3 \
     --api-key "$MOONSHOT_API_KEY" \
     --attributes '{"stage":"trial"}' \
     --gateway prod-gateway
    
   $ dome models add kimi-fireworks \
     --provider fireworks \
     --model accounts/fireworks/models/kimi-k3 \
     --api-key "$FIREWORKS_API_KEY" \
     --attributes '{"stage":"trial"}' \
     --gateway prod-gateway

2. 02

   Pool them

   Moonshot answers first and Fireworks second. The evaluation agents call kimi-k3, the pool.

   $ dome models pool create kimi-k3 \
     --failover-max all --gateway prod-gateway
    
   $ dome models pool member add kimi-k3 kimi-moonshot --priority 0
   $ dome models pool member add kimi-k3 kimi-fireworks --priority 1

3. 03

   Point your agent at the Gateway

   Code written for Moonshot's API runs against the Gateway once it carries the agent's Dome key.

   from openai import OpenAI
    
   client = OpenAI(base_url=f"{GATEWAY_URL}/v1", api_key=DOME_AGENT_KEY)
   client.chat.completions.create(
       model="kimi-k3",
       messages=[{"role": "user", "content": "Write tests for this module."}],
   )

Commands and rules tested against a Dome workspace on October 1, 2026. For anything about Moonshot AI itself, see Moonshot AI's documentation.

forbid (principal, action == Dome::Action::"llm:invoke", resource is Dome::LLMModel)
when { resource has stage && resource.stage == "trial" }
unless {
  principal has act_as &&
  principal.act_as.groups.contains("model-evals")
};