feat: add Model Serving connector and plugin

docs/docs/api/appkit/Interface.EndpointConfig.md

@@ -0,0 +1,21 @@
+# Interface: EndpointConfig
+
+## Properties
+
+### env
+
+```ts
+env: string;
+```
+
+Environment variable holding the endpoint name.
+
+***
+
+### servedModel?
+
+```ts
+optional servedModel: string;
+```
+
+Target a specific served model (bypasses traffic routing).

docs/docs/api/appkit/Interface.ServingEndpointEntry.md

@@ -0,0 +1,27 @@
+# Interface: ServingEndpointEntry
+
+Shape of a single registry entry.
+
+## Properties
+
+### chunk
+
+```ts
+chunk: unknown;
+```
+
+***
+
+### request
+
+```ts
+request: Record<string, unknown>;
+```
+
+***
+
+### response
+
+```ts
+response: unknown;
+```

docs/docs/api/appkit/Interface.ServingEndpointRegistry.md

@@ -0,0 +1,5 @@
+# Interface: ServingEndpointRegistry
+
+Registry interface for serving endpoint type generation.
+Empty by default — augmented by the Vite type generator's `.d.ts` output via module augmentation.
+When populated, provides autocomplete for alias names and typed request/response/chunk per endpoint.

docs/docs/api/appkit/TypeAlias.ServingFactory.md

@@ -0,0 +1,19 @@
+# Type Alias: ServingFactory
+
+```ts
+type ServingFactory = keyof ServingEndpointRegistry extends never ? (alias?: string) => ServingEndpointMethods : <K>(alias: K) => ServingEndpointMethods<ServingEndpointRegistry[K]["request"], ServingEndpointRegistry[K]["response"], ServingEndpointRegistry[K]["chunk"]>;
+```
+
+Factory function returned by `AppKit.serving`.
+
+This is a conditional type that adapts based on whether `ServingEndpointRegistry`
+has been populated via module augmentation (generated by `appKitServingTypesPlugin()`):
+
+- **Registry empty (default):** `(alias?: string) => ServingEndpointMethods` —
+  accepts any alias string with untyped request/response/chunk.
+- **Registry populated:** `<K>(alias: K) => ServingEndpointMethods<...>` —
+  restricts `alias` to known endpoint keys and infers typed request/response/chunk
+  from the registry entry.
+
+Run `appKitServingTypesPlugin()` in your Vite config to generate the registry
+augmentation and enable full type safety.

docs/docs/api/appkit/index.md

@@ -33,6 +33,7 @@ plugin architecture, and React integration.
 | [BasePluginConfig](Interface.BasePluginConfig.md) | Base configuration interface for AppKit plugins |
 | [CacheConfig](Interface.CacheConfig.md) | Configuration for the CacheInterceptor. Controls TTL, size limits, storage backend, and probabilistic cleanup. |
 | [DatabaseCredential](Interface.DatabaseCredential.md) | Database credentials with OAuth token for Postgres connection |
+| [EndpointConfig](Interface.EndpointConfig.md) | - |
 | [GenerateDatabaseCredentialRequest](Interface.GenerateDatabaseCredentialRequest.md) | Request parameters for generating database OAuth credentials |
 | [ITelemetry](Interface.ITelemetry.md) | Plugin-facing interface for OpenTelemetry instrumentation. Provides a thin abstraction over OpenTelemetry APIs for plugins. |
 | [LakebasePoolConfig](Interface.LakebasePoolConfig.md) | Configuration for creating a Lakebase connection pool |
@@ -42,6 +43,8 @@ plugin architecture, and React integration.
 | [ResourceEntry](Interface.ResourceEntry.md) | Internal representation of a resource in the registry. Extends ResourceRequirement with resolution state and plugin ownership. |
 | [ResourceFieldEntry](Interface.ResourceFieldEntry.md) | Defines a single field for a resource. Each field has its own environment variable and optional description. Single-value types use one key (e.g. id); multi-value types (database, secret) use multiple (e.g. instance_name, database_name or scope, key). |
 | [ResourceRequirement](Interface.ResourceRequirement.md) | Declares a resource requirement for a plugin. Can be defined statically in a manifest or dynamically via getResourceRequirements(). Narrows the generated base: type → ResourceType enum, permission → ResourcePermission union. |
+| [ServingEndpointEntry](Interface.ServingEndpointEntry.md) | Shape of a single registry entry. |
+| [ServingEndpointRegistry](Interface.ServingEndpointRegistry.md) | Registry interface for serving endpoint type generation. Empty by default — augmented by the Vite type generator's `.d.ts` output via module augmentation. When populated, provides autocomplete for alias names and typed request/response/chunk per endpoint. |
 | [StreamExecutionSettings](Interface.StreamExecutionSettings.md) | Execution settings for streaming endpoints. Extends PluginExecutionSettings with SSE stream configuration. |
 | [TelemetryConfig](Interface.TelemetryConfig.md) | OpenTelemetry configuration for AppKit applications |
 | [ValidationResult](Interface.ValidationResult.md) | Result of validating all registered resources against the environment. |
@@ -54,6 +57,7 @@ plugin architecture, and React integration.
 | [IAppRouter](TypeAlias.IAppRouter.md) | Express router type for plugin route registration |
 | [PluginData](TypeAlias.PluginData.md) | Tuple of plugin class, config, and name. Created by `toPlugin()` and passed to `createApp()`. |
 | [ResourcePermission](TypeAlias.ResourcePermission.md) | Union of all possible permission levels across all resource types. |
+| [ServingFactory](TypeAlias.ServingFactory.md) | Factory function returned by `AppKit.serving`. |
 | [ToPlugin](TypeAlias.ToPlugin.md) | Factory function type returned by `toPlugin()`. Accepts optional config and returns a PluginData tuple. |
 
 ## Variables

docs/docs/api/appkit/typedoc-sidebar.ts

@@ -97,6 +97,11 @@ const typedocSidebar: SidebarsConfig = {
           id: "api/appkit/Interface.DatabaseCredential",
           label: "DatabaseCredential"
         },
+        {
+          type: "doc",
+          id: "api/appkit/Interface.EndpointConfig",
+          label: "EndpointConfig"
+        },
         {
           type: "doc",
           id: "api/appkit/Interface.GenerateDatabaseCredentialRequest",
@@ -142,6 +147,16 @@ const typedocSidebar: SidebarsConfig = {
           id: "api/appkit/Interface.ResourceRequirement",
           label: "ResourceRequirement"
         },
+        {
+          type: "doc",
+          id: "api/appkit/Interface.ServingEndpointEntry",
+          label: "ServingEndpointEntry"
+        },
+        {
+          type: "doc",
+          id: "api/appkit/Interface.ServingEndpointRegistry",
+          label: "ServingEndpointRegistry"
+        },
         {
           type: "doc",
           id: "api/appkit/Interface.StreamExecutionSettings",
@@ -183,6 +198,11 @@ const typedocSidebar: SidebarsConfig = {
           id: "api/appkit/TypeAlias.ResourcePermission",
           label: "ResourcePermission"
         },
+        {
+          type: "doc",
+          id: "api/appkit/TypeAlias.ServingFactory",
+          label: "ServingFactory"
+        },
         {
           type: "doc",
           id: "api/appkit/TypeAlias.ToPlugin",

docs/static/appkit-ui/styles.gen.css

@@ -831,9 +831,6 @@
   .max-w-\[calc\(100\%-2rem\)\] {
     max-width: calc(100% - 2rem);
   }
-  .max-w-full {
-    max-width: 100%;
-  }
   .max-w-max {
     max-width: max-content;
   }
@@ -4514,6 +4511,11 @@
       width: calc(var(--spacing) * 5);
     }
   }
+  .\[\&_\[data-slot\=scroll-area-viewport\]\>div\]\:\!block {
+    & [data-slot=scroll-area-viewport]>div {
+      display: block !important;
+    }
+  }
   .\[\&_a\]\:underline {
     & a {
       text-decoration-line: underline;
@@ -4637,11 +4639,26 @@
       color: var(--muted-foreground);
     }
   }
+  .\[\&_table\]\:block {
+    & table {
+      display: block;
+    }
+  }
+  .\[\&_table\]\:max-w-full {
+    & table {
+      max-width: 100%;
+    }
+  }
   .\[\&_table\]\:border-collapse {
     & table {
       border-collapse: collapse;
     }
   }
+  .\[\&_table\]\:overflow-x-auto {
+    & table {
+      overflow-x: auto;
+    }
+  }
   .\[\&_table\]\:text-xs {
     & table {
       font-size: var(--text-xs);
@@ -4851,6 +4868,11 @@
       width: 100%;
     }
   }
+  .\[\&\>\*\]\:min-w-0 {
+    &>* {
+      min-width: calc(var(--spacing) * 0);
+    }
+  }
   .\[\&\>\*\]\:focus-visible\:relative {
     &>* {
       &:focus-visible {

packages/appkit/src/connectors/serving/client.ts

@@ -0,0 +1,163 @@
+import type { serving } from "@databricks/sdk-experimental";
+import { ApiError, type WorkspaceClient } from "@databricks/sdk-experimental";
+import { createLogger } from "../../logging/logger";
+import type { ServingStreamOptions } from "./types";
+
+const logger = createLogger("connectors:serving");
+
+/**
+ * Maps upstream Databricks error status codes to appropriate proxy responses.
+ * Used for raw API responses where the SDK doesn't handle errors automatically.
+ */
+function mapUpstreamError(status: number, body: string): ApiError {
+  const safeMessage = body.length > 500 ? `${body.slice(0, 500)}...` : body;
+
+  let parsed: { message?: string; error?: string } = {};
+  try {
+    parsed = JSON.parse(body);
+  } catch {
+    // body is not JSON
+  }
+
+  const message = parsed.message || parsed.error || safeMessage;
+
+  switch (true) {
+    case status === 400:
+      return new ApiError(message, "BAD_REQUEST", 400, undefined, []);
+    case status === 401 || status === 403:
+      logger.warn("Authentication failure from serving endpoint: %s", message);
+      return new ApiError(message, "AUTH_FAILURE", status, undefined, []);
+    case status === 404:
+      return new ApiError(message, "NOT_FOUND", 404, undefined, []);
+    case status === 429:
+      return new ApiError(message, "RATE_LIMITED", 429, undefined, []);
+    case status === 503:
+      return new ApiError(
+        "Endpoint loading, retry shortly",
+        "SERVICE_UNAVAILABLE",
+        503,
+        undefined,
+        [],
+      );
+    case status >= 500:
+      return new ApiError(message, "BAD_GATEWAY", 502, undefined, []);
+    default:
+      return new ApiError(message, "UNKNOWN", status, undefined, []);
+  }
+}
+
+/**
+ * Invokes a serving endpoint using the SDK's high-level query API.
+ * Returns a typed QueryEndpointResponse.
+ */
+export async function invoke(
+  client: WorkspaceClient,
+  endpointName: string,
+  body: Record<string, unknown>,
+): Promise<serving.QueryEndpointResponse> {
+  // Strip `stream` from the body — the connector controls this
+  const { stream: _stream, ...cleanBody } = body;
+
+  logger.debug("Invoking endpoint %s", endpointName);
+
+  return client.servingEndpoints.query({
+    name: endpointName,
+    ...cleanBody,
+  } as serving.QueryEndpointInput);
+}
+
+/**
+ * Invokes a serving endpoint with streaming enabled.
+ * Yields parsed JSON chunks from the SSE response.
+ *
+ * Uses the SDK's low-level `apiClient.request({ raw: true })` because
+ * the high-level `servingEndpoints.query()` returns `Promise<QueryEndpointResponse>`
+ * and does not support SSE streaming.
+ */
+export async function* stream(
+  client: WorkspaceClient,
+  endpointName: string,
+  body: Record<string, unknown>,
+  options?: ServingStreamOptions,
+): AsyncGenerator<unknown> {
+  // Strip any user-provided `stream` and inject `stream: true`
+  const { stream: _stream, ...cleanBody } = body;
+
+  logger.debug("Streaming from endpoint %s", endpointName);
+
+  const response = (await client.apiClient.request({
+    path: `/serving-endpoints/${encodeURIComponent(endpointName)}/invocations`,
+    method: "POST",
+    headers: new Headers({
+      "Content-Type": "application/json",
+      Accept: "text/event-stream",
+    }),
+    payload: { ...cleanBody, stream: true },
+    raw: true,
+  })) as { contents: ReadableStream<Uint8Array> };
+
+  if (!response.contents) {
+    throw new Error("Response body is null — streaming not supported");
+  }
+
+  const reader = response.contents.getReader();
+  const decoder = new TextDecoder();
+  let buffer = "";
+  const MAX_BUFFER_SIZE = 1024 * 1024; // 1 MB
+
+  try {
+    while (true) {
+      if (options?.signal?.aborted) break;
+
+      const { done, value } = await reader.read();
+      if (done) break;
+
+      buffer += decoder.decode(value, { stream: true });
+
+      if (buffer.length > MAX_BUFFER_SIZE) {
+        throw new Error(
+          `Stream buffer exceeded ${MAX_BUFFER_SIZE} bytes — possible non-SSE response`,
+        );
+      }
+
+      // Process complete lines from the buffer
+      const lines = buffer.split("\n");
+      // Keep the last (potentially incomplete) line in the buffer
+      buffer = lines.pop() ?? "";
+
+      for (const line of lines) {
+        const trimmed = line.trim();
+        // Per SSE spec: empty lines are event delimiters,
+        // lines starting with ":" are comments (often used as heartbeats).
+        if (!trimmed || trimmed.startsWith(":")) continue;
+        if (trimmed === "data: [DONE]") return;
+
+        if (trimmed.startsWith("data: ")) {
+          const jsonStr = trimmed.slice(6);
+          try {
+            yield JSON.parse(jsonStr);
+          } catch {
+            logger.warn("Failed to parse streaming chunk: %s", jsonStr);
+          }
+        }
+      }
+    }
+
+    // Process any remaining data in the buffer
+    if (buffer.trim() && !options?.signal?.aborted) {
+      const trimmed = buffer.trim();
+      if (trimmed.startsWith("data: ") && trimmed !== "data: [DONE]") {
+        try {
+          yield JSON.parse(trimmed.slice(6));
+        } catch {
+          logger.warn("Failed to parse final streaming chunk: %s", trimmed);
+        }
+      }
+    }
+  } finally {
+    if (options?.signal?.aborted) {
+      reader.cancel().catch(() => {});
+    }
+    reader.releaseLock();
+  }
+}