Kimi and Dome
Run Kimi K3 from three providers under one name. Kimi is an open-weight model built for coding. Three places serve it: Moonshot AI, which makes it, plus Together AI and Fireworks AI. Dome puts all three under kimi-k3, gives it to people with the engineer role, and caps the month.
Dome provides model brokeringand routing for Kimi.
Moonshot first, then two hosts
Pool Moonshot's own API with Together AI and Fireworks AI under kimi-k3. When Moonshot fails, the call moves down the list.
Kimi for engineers
Tag Kimi connections with a family attribute. A rule refuses them unless the person the agent acts for holds the engineer role.
A ceiling on spend
A $400 monthly quota on the pool counts all three hosts. Kimi stops there; the agent's other models don't.
Kimi with failoverin five steps.
Moonshot plus two hosts, one kimi-k3 pool, a role rule and a monthly cap.
Connect Moonshot AI
Moonshot's API is OpenAI-compatible at api.moonshot.ai. Dome keeps its key and sends it as a bearer token.
shelldome models add kimi-moonshot \ --provider moonshot \ --model kimi-k3 \ --api-key "$MOONSHOT_API_KEY" \ --attributes '{"family":"kimi"}' \ --gateway prod-gatewayConnect Together AI and Fireworks AI
Each host serves the same model under its own id, so tag them the same way.
shelldome models add kimi-together \ --provider together \ --model moonshotai/Kimi-K3 \ --api-key "$TOGETHER_API_KEY" \ --attributes '{"family":"kimi"}' \ --gateway prod-gateway dome models add kimi-fireworks \ --provider fireworks \ --model accounts/fireworks/models/kimi-k3 \ --api-key "$FIREWORKS_API_KEY" \ --attributes '{"family":"kimi"}' \ --gateway prod-gatewayPool them for failover
Moonshot first, then Together AI, then Fireworks AI. A coding agent asks for kimi-k3 and gets whichever answers.
shelldome models pool create kimi-k3 \ --failover-max all --gateway prod-gateway dome models pool member add kimi-k3 kimi-moonshot --priority 0dome models pool member add kimi-k3 kimi-together --priority 1dome models pool member add kimi-k3 kimi-fireworks --priority 2Point your agent at the Gateway
Swap the base URL for the Gateway and the key for the agent's Dome key. Nothing else in the client changes.
pythonfrom 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": "Refactor this module to remove the global state."}],)Cap the spend
The cap is on the pool, so moving between hosts doesn't reset it.
shelldome quotas set --subject pool --pool kimi-k3 \ --unit provider_usd --limit 400 --window monthly
Kimifor engineers.
Scoped to one agent, this checks the role in the caller's identity token. Engineers reach Kimi through it, and anyone else is refused, on Kimi only.
cedar
forbid (principal, action == Dome::Action::"llm:invoke", resource is Dome::LLMModel)when { resource has family && resource.family == "kimi" }unless { principal has act_as && principal.act_as.roles.contains("engineer") };One call,two outcomes.
Switch the caller or the argument and watch the same call decide differently, with every decision landing in audit.
Acting for
agent coding-agent · acting as d.chen
llm:invoke(model: "kimi-k3")
- Agentcoding-agent is registered and active
- Callerd.chen verified, roles: engineer
- RuleKimi is open to engineers
- Quota$142 of $400 this month
DecisionAllowed
Bringing it alltogether.
Connecting Kimi to registered agents, tools, and identity in Dome completes a governed agent application.
Acting for
Control point
GatewayEvery call decided and audited
RulesGuardsQuotas
Models
Common questionsabout Kimi.
More of whatDome works with.
Provider
Anthropic
The Claude API behind the Model Broker: the key held in Dome, every call authorized, metered and audited.
MCP server
GitHub
The GitHub MCP server behind the Tool Gateway, with rules that decide each call on its owner and repository.
Identity
Okta
Okta tokens verified on every agent call, so rules and audit name the person each agent acted for.
Runtime
LangGraph
LangGraph agents with their model calls on the Model Broker and their MCP tools on the Tool Gateway.
Client
Claude Code
Claude Code on a Dome Gateway with per-developer sign-in, rules on every tool call, and audit by name.
Agent service
TinyFish
TinyFish's web agents behind the Tool Gateway, with rules that decide each run on the site it targets.
Talk with ourFDE team.
Or start free, and register your first agent in minutes. Connect it through a gateway, give it rules, and see every call it makes in one audit trail.
Talk with our FDE team
Our engineers design, build and run governed agents for the part of your business you want to change. You see what they deliver.
AI models for enterprise agents: failover, rules and quotas
Every model your agents call goes through the Model Broker. Pool providers for failover, decide who may call each model, and cap what it costs.