Unity Gateway
Use the parent databricks-core skill for CLI authentication and profile selection.
Unity Gateway provides Unity Catalog securables for governed access to AI services. The CLI
command group keeps the earlier product name: databricks ai-gateway --help.
Choose the Correct Gateway
| Request |
Use |
| Unity Gateway model service, MCP service, or model provider service |
This skill and databricks ai-gateway |
AI Gateway configuration on a Model Serving endpoint, including put-ai-gateway |
databricks-model-serving and databricks serving-endpoints |
| Lakeflow Connect ingestion gateway |
databricks-lakeflow-connect |
Treat Unity AI Gateway as an earlier name for Unity Gateway when the resource is a
Unity Catalog securable. Do not translate legacy Model Serving AI Gateway resources into
Unity Gateway resources unless the user explicitly requests a migration.
CLI Workflow
Use the Databricks CLI for resource lifecycle and permission management. Do not substitute
direct REST calls or a Databricks SDK for those operations. Python clients are supported
for querying model and model provider services; read Query services.
An explicit request for Unity Gateway, Unity AI Gateway, or
databricks ai-gateway belongs to this skill. Do not switch to
databricks-model-serving merely because the request involves a model service or a
workspace. Use that skill only for a legacy per-endpoint AI Gateway operation or when a
provisioned-throughput destination requires Model Serving endpoint details.
Verify the CLI meets the minimum version:
databricks --version
Inspect the command and the relevant operation before constructing a payload:
databricks ai-gateway --help
databricks ai-gateway <operation> --help
Pass create and update configuration through --json. Use an explicit profile when
profile-based authentication is required:
databricks ai-gateway <operation> --json @payload.json --profile <PROFILE>
Do not invent JSON fields from similarly named legacy APIs. Build the payload from the
installed CLI help and the Unity Gateway documentation for the selected service type.
Service Types
| Service type |
CLI operations |
| Model service |
create-model-service, get-model-service, list-model-services, update-model-service, delete-model-service; read Model services |
| MCP service |
create-mcp-service, get-mcp-service, list-mcp-services, update-mcp-service, delete-mcp-service; read MCP services |
| Model provider service |
create-model-provider-service, get-model-provider-service, list-model-provider-services, update-model-provider-service, delete-model-provider-service; read Model provider services |
Querying and permissions
- Before writing Python that invokes a model or model provider service, read references/querying.md.
- Before checking, granting, or revoking Unity Gateway access, read references/permissions.md.
- Before any model-service operation, read references/model-services.md. It contains the current resource-name conventions, JSON payload fields, required-input gate, and lifecycle commands.
- Before any MCP-service operation, read references/mcp-services.md for the equivalent MCP-specific contract.
- Before any model-provider-service operation, read references/model-provider-services.md.
- Before setting up a spend budget or spend alert on Unity Gateway usage, read references/budgets.md.
- Before any account-scoped operation (account groups as grant principals, account budgets), read references/account-access.md.
Do not fetch public documentation for fields already covered by these references. Consult the
authoritative documentation only when a required field is absent from the reference or the
user explicitly asks for the latest documentation. Do not assign a Beta or preview status
unless the installed CLI help or current documentation explicitly does so.
Migration from Model Serving to Unity Gateway
When migrating from workspace-scoped Model Serving endpoints to Unity Catalog-scoped Unity Gateway services, read references/model-serving-migration.md.
Authoritative Documentation
1---2name: databricks-unity-gateway3description: >-4 Use when querying or calling pay-per-token foundation models, or when5 creating, updating, listing, or querying Unity Gateway (also called Unity AI6 Gateway) services, including model provider services, MCP services, system.ai7 services, and three-part Unity Catalog service names (catalog.schema.service).8 Also use for managing spend budgets and usage alerts on Unity Gateway, and for9 migrating from legacy AI Gateway. Also use when a workload with Model Serving10 endpoint permissions (for example CAN_QUERY) gets PERMISSION_DENIED querying a11 three-part model name, since serving-endpoint access does not grant access to a12 Unity Gateway service. Not for configuring legacy AI Gateway on Model Serving13 endpoints; use databricks-model-serving for that.14compatibility: Requires databricks CLI (>= v1.11.0)15metadata:16 source: https://skillmd.com/skills/databricks/databricks-unity-gateway17 version: "0.1.0"18parent: databricks-core19---2021# Unity Gateway2223Use the parent `databricks-core` skill for CLI authentication and profile selection.2425Unity Gateway provides Unity Catalog securables for governed access to AI services. The CLI26command group keeps the earlier product name: `databricks ai-gateway --help`.2728## Choose the Correct Gateway2930| Request | Use |31|---|---|32| Unity Gateway model service, MCP service, or model provider service | This skill and `databricks ai-gateway` |33| AI Gateway configuration on a Model Serving endpoint, including `put-ai-gateway` | `databricks-model-serving` and `databricks serving-endpoints` |34| Lakeflow Connect ingestion gateway | `databricks-lakeflow-connect` |3536Treat **Unity AI Gateway** as an earlier name for **Unity Gateway** when the resource is a37Unity Catalog securable. Do not translate legacy Model Serving AI Gateway resources into38Unity Gateway resources unless the user explicitly requests a migration.3940## CLI Workflow4142Use the Databricks CLI for resource lifecycle and permission management. Do not substitute43direct REST calls or a Databricks SDK for those operations. Python clients are supported44for querying model and model provider services; read [Query services](references/querying.md).4546An explicit request for **Unity Gateway**, **Unity AI Gateway**, or47`databricks ai-gateway` belongs to this skill. Do not switch to48`databricks-model-serving` merely because the request involves a model service or a49workspace. Use that skill only for a legacy per-endpoint AI Gateway operation or when a50provisioned-throughput destination requires Model Serving endpoint details.51521. Verify the CLI meets the minimum version:5354 ```bash55 databricks --version56 ```57582. Inspect the command and the relevant operation before constructing a payload:5960 ```bash61 databricks ai-gateway --help62 databricks ai-gateway <operation> --help63 ```64653. Pass create and update configuration through `--json`. Use an explicit profile when66 profile-based authentication is required:6768 ```bash69 databricks ai-gateway <operation> --json @payload.json --profile <PROFILE>70 ```7172Do not invent JSON fields from similarly named legacy APIs. Build the payload from the73installed CLI help and the Unity Gateway documentation for the selected service type.7475## Service Types7677| Service type | CLI operations |78|---|---|79| Model service | `create-model-service`, `get-model-service`, `list-model-services`, `update-model-service`, `delete-model-service`; read [Model services](references/model-services.md) |80| MCP service | `create-mcp-service`, `get-mcp-service`, `list-mcp-services`, `update-mcp-service`, `delete-mcp-service`; read [MCP services](references/mcp-services.md) |81| Model provider service | `create-model-provider-service`, `get-model-provider-service`, `list-model-provider-services`, `update-model-provider-service`, `delete-model-provider-service`; read [Model provider services](references/model-provider-services.md) |8283## Querying and permissions8485- Before writing Python that invokes a model or model provider service, read [references/querying.md](references/querying.md).86- Before checking, granting, or revoking Unity Gateway access, read [references/permissions.md](references/permissions.md).87- Before any model-service operation, read [references/model-services.md](references/model-services.md). It contains the current resource-name conventions, JSON payload fields, required-input gate, and lifecycle commands.88- Before any MCP-service operation, read [references/mcp-services.md](references/mcp-services.md) for the equivalent MCP-specific contract.89- Before any model-provider-service operation, read [references/model-provider-services.md](references/model-provider-services.md).90- Before setting up a spend budget or spend alert on Unity Gateway usage, read [references/budgets.md](references/budgets.md).91- Before any account-scoped operation (account groups as grant principals, account budgets), read [references/account-access.md](references/account-access.md).9293Do not fetch public documentation for fields already covered by these references. Consult the94authoritative documentation only when a required field is absent from the reference or the95user explicitly asks for the latest documentation. Do not assign a Beta or preview status96unless the installed CLI help or current documentation explicitly does so.9798## Migration from Model Serving to Unity Gateway99100When migrating from workspace-scoped Model Serving endpoints to Unity Catalog-scoped Unity Gateway services, read [references/model-serving-migration.md](references/model-serving-migration.md).101102## Authoritative Documentation103104- [Unity Gateway overview](https://docs.databricks.com/aws/en/ai-gateway/)105- [Unity Gateway release notes](https://docs.databricks.com/aws/en/release-notes/unity-gateway/)106- [Unity Gateway developer documentation](https://developers.databricks.com/docs/agents/ai-gateway)107- [Model service API reference](https://docs.databricks.com/api/ai-gateway/v1/model-service)108- [MCP service API reference](https://docs.databricks.com/api/ai-gateway/v1/mcp-service)109- [Model provider service API reference](https://docs.databricks.com/api/ai-gateway/v1/model-provider-service)110- [Query model services](https://docs.databricks.com/aws/en/ai-gateway/query-model-services)111- [Open Responses specification](https://www.openresponses.org/specification)112- [Open Responses provider behavior on Databricks](https://docs.databricks.com/aws/en/machine-learning/model-serving/query-open-responses-models)113- [Query model provider services](https://docs.databricks.com/aws/en/ai-gateway/query-model-provider-services)114- [Model service requirements and grants](https://docs.databricks.com/ai-gateway/model-services)115- [Model provider service governance](https://docs.databricks.com/aws/en/ai-gateway/govern-model-provider-services)116- [Routing and fallbacks](https://docs.databricks.com/aws/en/ai-gateway/configure-traffic-splitting)117- [Rate limits](https://docs.databricks.com/aws/en/ai-gateway/rate-limits)118- [Register an external MCP server](https://docs.databricks.com/aws/en/ai-gateway/register-mcp-service)