Security & deployment

Architecture

Where Miso sits and what it controls.

Customer-cloud deployment

The Miso gateway and policy store are deployed within your cloud account, so model credentials, governance configuration, and audit records remain under your organization's administrative control.

Compatible with existing applications

Applications connect through an OpenAI-compatible API rather than integrating separately with each model provider. Existing AI applications are pointed at Miso; their code stays as it is.

Designed for enterprise operations

Miso works with your infrastructure, security, and privacy teams through deployment, initial policy configuration, and production rollout, and maintains the software after.

A deliberately small footprint

The core deployment consists of an application container, a PostgreSQL database, and the cloud services you already run: identity, secrets, networking, logging. This keeps the infrastructure a health system has to introduce and operate to a minimum.

Application containerThe gateway: endpoint, policy engine, identifier detection, budgets, audit writer

PostgreSQLPolicy store and append-only audit tables, in your database service

Existing cloud servicesIdentity provider, secrets manager, load balancer, log pipeline

Security boundary

What remains inside, what may leave, where credentials and records reside.

Stays in your environment

Miso policy and configuration
Application identities and credentials
Model-provider credentials
Approval history and route requests
Audit records
Governance metadata: registries, obligations, budgets

May leave your environment

Only traffic sent to model providers through routes your organization has authorized.

Identifiers the route names are replaced before the request leaves. The provider that receives it is on your registry with an agreement in force. The record of the request stays inside.

Miso does not create a second, external repository for the governance data required to operate the gateway.

Data in transitTLS between callers and Miso, and between Miso and providers

Data at restYour PostgreSQL and your key management; Miso stores hashes, identifier types and counts, and the record fields. Prompt and response text are off by default.

AccessConsole sign-in through your OIDC identity provider. The console reads through a read-only database role. Audit tables accept inserts only, enforced by a database role.

RetentionSix years by default, set by you

Failure modeIf the gateway is unreachable, callers receive an error and no request is sent

Enterprise integration

Miso is a control point inside an architecture you already own. It does not replace your cloud, identity, or security stack.

Existing systemMiso's role

AWS, Azure, or Google CloudRuns within it, as a container and a database in your account

Your identity provider (OIDC)Uses identity from it for console access and approvals

Your log pipeline or SIEMSends operational and security events to it

Anthropic, Azure OpenAI, AWS Bedrock, Google VertexGoverns access to them under your agreements

GRC and contract systemsEnforces the decisions recorded in them; exports the record back

Internal applications and AI vendorsGives them governed model access through one endpoint

Production rollout

How Miso and your teams get the first applications governed.

A narrow integration surface

Miso fits into existing cloud and application environments without requiring teams to rebuild their AI applications. Applications connect through an OpenAI-compatible endpoint. Miso works with the health system to establish the surrounding identity, network, provider, and governance configuration.

Establish the environment

Deploy Miso within the health system's cloud boundary and connect identity, networking, and logging.

Connect approved model providers

Configure the model accounts and the contractual relationships the organization has already approved.

Encode initial governance policies

Translate approved use cases, data classes, models, and agreements into enforceable routes. The policy is validated before it is deployed.

Bring applications onto Miso

Move selected internal applications or vendors to governed endpoints and expand from there.

Miso

Drafts the policy with your privacy and security teams
Checks each provider's current published terms against its agreement
Onboards the first application and the first vendor with you
Reviews the first month's record with you and tunes the routes

Your organization

Names a privacy owner and a security owner for approvals
Operates the container and the database in your account
Decides which applications go first
Holds the provider agreements and credentials

Operations

Your environment. A supported Miso deployment.

Miso maintains the software, delivers updates, and supports the deployment as your AI environment evolves. Your organization controls the infrastructure it runs on.

UpdatesMiso ships versioned releases. You apply them on your schedule; in a Miso-operated deployment, Miso applies them under the support agreement.

Configuration changesPolicy changes are validated before deployment and recorded with a date and a hash. Route requests and approvals go through the console and land in the policy.

MonitoringA health endpoint, and operational and security events to your log pipeline.

AvailabilityThe gateway is stateless. Run two or more instances behind your load balancer.

RecoveryBackups are your database's backups. A gateway instance is recreated from the release image and the policy.

SupportNamed contacts at Miso through rollout and after, under a support agreement.

AttestationsAsk us for the current status of third-party assessments. Architecture, test results, and the dependency list are available to prospective customers under NDA.

Miso runs where your AI does.