Managed services

Request PostgreSQL + PgBouncer, MLflow or LiteLLM; follow setup, obtain credentials and manage your instance.

PostgreSQL + PgBouncer, MLflow and LiteLLM are available as an internal preview. Open your managed services to request an instance and track its setup. These services do not currently charge your balance.

Setup

Choose the service, instance name and placement. Describe who needs access and the workload you expect. Do not put passwords or provider keys in the request.

An operator provisions the service through its owning platform, checks the connection and access boundary, and records the supported configuration. A request stays pending until that work is complete. “Ready · experimental” means the recorded connection check passed; it does not promise production availability or automatic failover.

The private platform and AWOW Azure are separate profiles. Selecting Azure does not move data out of that Azure boundary. The Azure profile needs AWOW's own deployment and qualification.

Connect

The instance page contains its endpoint, authentication method, network access, supported configuration, limitations and last verification time. Once ready, download your account's private connection credentials. Save that file in your application's secret store. Downloads are audited. PostgreSQL has its own database credentials. Managed MLflow and LiteLLM use a Browsergrid API key granted to their scoped endpoint.

For PostgreSQL, use the supplied CA with sslmode=verify-full. The pool mode matters: transaction pooling is unsuitable for unqualified session locks, LISTEN and other session-dependent operations. PgBouncer bounds the database connections used by application replicas; it does not distribute writes across database servers.

For MLflow, use the scoped tracking endpoint and a Browsergrid API key granted to this instance. Each instance states whether artifacts are shared across replicas. A profile backed by a local artifact volume remains a single-replica profile.

For LiteLLM, use the scoped gateway endpoint and your granted Browsergrid API key as its bearer token. Only the approved upstream routes are supported. Model execution belongs to the model provider; gateway cost and provider consumption are separate.

Manage an instance

Request credential rotation, a capacity change, restoration into a separate instance, suspension, resumption or retirement from its page. The request does not perform the operation immediately. The operator verifies the outcome and records a new status. Retirement is not automatic data deletion.

A shared internal instance is restricted to its declared internal account. It is not a separate tenant for every project using it. Request dedicated isolation when projects require separate access boundaries.

Cost and readiness

The instance page shows infrastructure expense attributed to its immutable workload identities by OpenCost and the accounting ledger. A partial collection is not a full-period total. Missing observations and shared database, storage or upstream expenses must be allocated separately; no recorded expense does not mean zero cost.

Each service keeps an operating history. MLflow qualification includes trace completeness, SQL connection budgets and consistent metadata/artifact restoration. LiteLLM qualification includes memory under real streaming load, scale-in and export completion. Capacity or image changes require fresh evidence for the changed profile.