Claude Opus and Dome

1. 01

   Connect Anthropic

   Tag the connection tier: frontier. The rule below reads that tag, and the Anthropic key stays in Dome's vault.

   $ dome models add opus-55-anthropic \
     --provider anthropic \
     --model claude-opus-5-5 \
     --api-key "$ANTHROPIC_API_KEY" \
     --attributes '{"tier":"frontier"}' \
     --gateway prod-gateway

2. 02

   Connect AWS Bedrock

   Add a Bedrock connection in the dashboard with your AWS keys and region. Static AWS keys can only be set there.

3. 03

   Pool them for failover

   Anthropic at priority 0, Bedrock at 1. Agents address the pool, so swapping a member later touches no agent code.

   $ dome models pool create claude-opus-5-5 \
     --failover-max all --gateway prod-gateway
    
   $ dome models pool member add claude-opus-5-5 opus-55-anthropic --priority 0
   $ dome models pool member add claude-opus-5-5 opus-55-bedrock --priority 1

4. 04

   Point your agent at the Gateway

   The research agent's client keeps its shape. Only the base URL and the key change, and the key is the agent's own.

   from openai import OpenAI
    
   client = OpenAI(base_url=f"{GATEWAY_URL}/v1", api_key=DOME_AGENT_KEY)
   client.chat.completions.create(
       model="claude-opus-5-5",
       messages=[{"role": "user", "content": "Summarize this incident."}],
   )

5. 05

   Cap the spend

   Quotas bind to the pool, a single connection, or each caller.

   $ dome quotas set --subject pool --pool claude-opus-5-5 \
     --unit provider_usd --limit 500 --window monthly
    
   $ dome quotas set --subject model --model opus-55-anthropic \
     --per-caller --unit tokens --limit 2000000 --window daily

Commands and rules tested against a Dome workspace on September 30, 2026. For anything about Claude Opus itself, see Anthropic's documentation.

forbid (principal, action == Dome::Action::"llm:invoke", resource is Dome::LLMModel)
when { resource has tier && resource.tier == "frontier" }
unless {
  principal has act_as &&
  principal.act_as.groups.contains("ml-research")
};