Dome Systems

Mistral and Dome

Run Mistral Large 3 from Mistral and NVIDIA under one name.

Mistral Large 3 has open weights, so Mistral's API is not the only place to run it. Pool it with NVIDIA NIM, and agents ask for mistral-large-3. A rule keeps work for EU staff on connections tagged eu.

PeopleAgentsDomePool: mistral-large-3CallersPeopleAgentshr-assistantAgentscontract-drafterAgentstranslatorGatewaysprod-gatewayMistralPriority 0, region: euNVIDIA NIMPriority 1, region: usAnthropicclaude-opus-5-5, no region tagRulesGuardsAuditsaudit-trail
hr-assistant→mistral-api· as l.moreauAllowed

How Dome helps

Dome provides model brokering and routing for Mistral

Mistral first, NVIDIA next

Mistral serves Large 3 as mistral-large-2512, and NVIDIA NIM serves the same open weights. When Mistral fails, the call goes to NVIDIA.

EU work on EU connections

Tag each connection with a region. When the person behind a call is in the EU group, a rule refuses anything not tagged eu.

Spend per person

Cap what one person's agents spend on models at $50 a month. It counts every agent that acts for them.

Get started

Mistral with failover in five steps

Tag Mistral and NVIDIA by region, pool them, add the EU rule, then cap spend per person.

  1. 01

    Connect Mistral

    The provider id is mistral. The region attribute is your label, and rules read it.

    $ dome models add mistral-api \
    --provider mistral \
    --model mistral-large-2512 \
    --api-key "$MISTRAL_API_KEY" \
    --attributes '{"family":"mistral","region":"eu"}' \
    --gateway prod-gateway
  2. 02

    Connect NVIDIA NIM

    The same open weights, under NVIDIA's model id.

    $ dome models add mistral-nvidia \
    --provider nvidia \
    --model mistralai/mistral-large-3-675b-instruct-2512 \
    --api-key "$NVIDIA_API_KEY" \
    --attributes '{"family":"mistral","region":"us"}' \
    --gateway prod-gateway
  3. 03

    Pool them for failover

    Mistral's API leads and NVIDIA follows. Agents only ever send mistral-large-3.

    $ dome models pool create mistral-large-3 \
    --failover-max all --gateway prod-gateway
     
    $ dome models pool member add mistral-large-3 mistral-api --priority 0
    $ dome models pool member add mistral-large-3 mistral-nvidia --priority 1
  4. 04

    Point your agent at the Gateway

    Mistral's API speaks the OpenAI format, so the agent's OpenAI client works once it points at the Gateway.

    from openai import OpenAI
     
    client = OpenAI(base_url=f"{GATEWAY_URL}/v1", api_key=DOME_AGENT_KEY)
    client.chat.completions.create(
    model="mistral-large-3",
    messages=[{"role": "user", "content": "Draft a reply to this leave request."}],
    )
  5. 05

    Cap spend per person

    An act-as quota follows the verified person, across every agent that acts for them.

    $ dome quotas set --subject act-as --act-as l.moreau \
    --dimension llm --unit provider_usd --limit 50 --window monthly

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

Rules

EU work stays on EU connections

Applied to an agent with --agent, this refuses any connection not tagged region: eu when the person the agent acts for is in the eu group. Everyone else can use either connection.

forbid (principal, action == Dome::Action::"llm:invoke", resource is Dome::LLMModel)
when { principal has act_as && principal.act_as.groups.contains("eu") }
unless { resource has region && resource.region == "eu" };

Try it

One call, two outcomes

Switch the caller or the argument and watch the same call decide differently. Every decision lands in audit.

Connection

agent hr-assistant · acting as l.moreau
llm:invoke(model: "mistral-api")
  1. Agenthr-assistant is registered and active
  2. Callerl.moreau verified, groups: eu
  3. RuleConnection is tagged region: eu
  4. Quota$21 of $50 this month
DecisionAllowed

Agent workflow

Bringing it together

Connecting Mistral to registered agents, tools, and identity in Dome completes a governed agent application.

Dome

Control point

Gateway

  • Rules
  • Guards
  • Quotas

Every call decided and audited

FAQ

Common questions

Which providers serve Mistral models through Dome?

Mistral's own API, NVIDIA NIM and Amazon Bedrock serve Mistral Large 3. Fireworks AI serves it on dedicated deployments.

Which Mistral models have open weights?

Mistral Large 3, Mistral Small 4 and the Ministral 3 models are released under Apache 2.0. Any host serving those weights can join a pool.

What happens when a new Mistral model ships?

Connect it on Mistral and on any host serving the open weights, with the same region tags, and swap the members.

Next steps

Talk with our FDE team

Our forward deployed engineers work with your platform team to get your agents into production and under control: the first one governed on your own systems, and a pattern your teams can repeat for every agent after it.