@homeflare/distilled-litellm 0.2.0 → 0.3.0

package/src/services/settings.ts

@@ -1,4 +1,4 @@
-// AUTO-GENERATED by scripts/generate.ts from .generated-specs (pinned litellm-openapi @ 1.100.0 — see README). Do not edit.
+// AUTO-GENERATED by scripts/generate.ts from .generated-specs (pinned litellm-openapi @ 1.103.0 — see README). Do not edit.
 import * as S from "@distilled.cloud/core/schema";
 import * as API from "@distilled.cloud/core/api";
 import * as C from "@distilled.cloud/core/category";
@@ -68,6 +68,102 @@ export const MCPSemanticFilterSettingsResponse = /*@__PURE__*/ S.suspend(() =>
   identifier: "MCPSemanticFilterSettingsResponse",
 }) as any as S.Schema<MCPSemanticFilterSettingsResponse>;
 
+export interface GetMcpToolSearchSettingsGetMcpToolSearchSettingsGetRequest {}
+export const GetMcpToolSearchSettingsGetMcpToolSearchSettingsGetRequest =
+  /*@__PURE__*/ S.suspend(() =>
+    S.Struct({}).pipe(
+      T.Http({
+        method: "GET",
+        uri: "/get/mcp_tool_search_settings",
+        code: 200,
+      }),
+    ),
+  ).annotate({
+    identifier: "GetMcpToolSearchSettingsGetMcpToolSearchSettingsGetRequest",
+  }) as any as S.Schema<GetMcpToolSearchSettingsGetMcpToolSearchSettingsGetRequest>;
+
+export type MCPToolSearchSettingsResponseFieldSchemaMap = {
+  [key: string]: unknown | undefined;
+};
+export const MCPToolSearchSettingsResponseFieldSchemaMap =
+  /*@__PURE__*/ S.Record(
+    S.String,
+    S.Unknown,
+  ) as any as S.Schema<MCPToolSearchSettingsResponseFieldSchemaMap>;
+
+export type MCPToolSearchSettingsResponseValuesMap = {
+  [key: string]: unknown | undefined;
+};
+export const MCPToolSearchSettingsResponseValuesMap = /*@__PURE__*/ S.Record(
+  S.String,
+  S.Unknown,
+) as any as S.Schema<MCPToolSearchSettingsResponseValuesMap>;
+
+/** Response model for native MCP tool search settings */
+export interface MCPToolSearchSettingsResponse {
+  field_schema: MCPToolSearchSettingsResponseFieldSchemaMap;
+  values: MCPToolSearchSettingsResponseValuesMap;
+}
+export const MCPToolSearchSettingsResponse = /*@__PURE__*/ S.suspend(() =>
+  S.Struct({
+    field_schema: MCPToolSearchSettingsResponseFieldSchemaMap,
+    values: MCPToolSearchSettingsResponseValuesMap,
+  }),
+).annotate({
+  identifier: "MCPToolSearchSettingsResponse",
+}) as any as S.Schema<MCPToolSearchSettingsResponse>;
+
+export interface GetWebsearchInterceptionSettingsGetWebsearchInterceptionSettingsGetRequest {}
+export const GetWebsearchInterceptionSettingsGetWebsearchInterceptionSettingsGetRequest =
+  /*@__PURE__*/ S.suspend(() =>
+    S.Struct({}).pipe(
+      T.Http({
+        method: "GET",
+        uri: "/get/websearch_interception_settings",
+        code: 200,
+      }),
+    ),
+  ).annotate({
+    identifier:
+      "GetWebsearchInterceptionSettingsGetWebsearchInterceptionSettingsGetRequest",
+  }) as any as S.Schema<GetWebsearchInterceptionSettingsGetWebsearchInterceptionSettingsGetRequest>;
+
+export type WebSearchInterceptionSettingsResponseFieldSchemaMap = {
+  [key: string]: unknown | undefined;
+};
+export const WebSearchInterceptionSettingsResponseFieldSchemaMap =
+  /*@__PURE__*/ S.Record(
+    S.String,
+    S.Unknown,
+  ) as any as S.Schema<WebSearchInterceptionSettingsResponseFieldSchemaMap>;
+
+export type WebSearchInterceptionSettingsResponseValuesMap = {
+  [key: string]: unknown | undefined;
+};
+export const WebSearchInterceptionSettingsResponseValuesMap =
+  /*@__PURE__*/ S.Record(
+    S.String,
+    S.Unknown,
+  ) as any as S.Schema<WebSearchInterceptionSettingsResponseValuesMap>;
+
+/** Response model for web search interception settings */
+export interface WebSearchInterceptionSettingsResponse {
+  /** Whether the process answering this request has the interception callback registered. Read-only: it reports what is running here, while values.enabled is the cluster-wide setting, and the two disagree while a pod is still applying a change or failed to apply it. */
+  active_on_this_pod?: boolean;
+  field_schema: WebSearchInterceptionSettingsResponseFieldSchemaMap;
+  values: WebSearchInterceptionSettingsResponseValuesMap;
+}
+export const WebSearchInterceptionSettingsResponse = /*@__PURE__*/ S.suspend(
+  () =>
+    S.Struct({
+      active_on_this_pod: S.optional(S.Boolean),
+      field_schema: WebSearchInterceptionSettingsResponseFieldSchemaMap,
+      values: WebSearchInterceptionSettingsResponseValuesMap,
+    }),
+).annotate({
+  identifier: "WebSearchInterceptionSettingsResponse",
+}) as any as S.Schema<WebSearchInterceptionSettingsResponse>;
+
 export interface UpdateMcpSemanticFilterSettingsUpdateMcpSemanticFilterSettingsPatchRequest {
   /** Embedding model to use for semantic similarity (e.g., 'text-embedding-3-small', 'text-embedding-ada-002') */
   embedding_model?: string;
@@ -110,6 +206,118 @@ export const UpdateMcpSemanticFilterSettingsUpdateMcpSemanticFilterSettingsPatch
       "UpdateMcpSemanticFilterSettingsUpdateMcpSemanticFilterSettingsPatchResponse",
   }) as any as S.Schema<UpdateMcpSemanticFilterSettingsUpdateMcpSemanticFilterSettingsPatchResponse>;
 
+/** Tool names always returned first when the caller can access them, e.g. `my_server-get_rates`. */
+export type UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchRequestCoreToolsList =
+  Array<string>;
+export const UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchRequestCoreToolsList =
+  /*@__PURE__*/ S.Array(
+    S.String,
+  ) as any as S.Schema<UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchRequestCoreToolsList>;
+
+export interface UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchRequest {
+  /** Tool names always returned first when the caller can access them, e.g. `my_server-get_rates`. */
+  core_tools?: UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchRequestCoreToolsList;
+  /** Embedding model from model_list used to rank tools by meaning. Unset keeps keyword matching. */
+  embedding_model?: string | null;
+  /** Lowest cosine similarity a tool needs to appear in semantic results (0.0 = no cutoff). */
+  similarity_threshold?: number;
+  /** Most ranked tools a search returns. A smaller top_k in the tool call wins. Core tools do not count. */
+  top_k?: number;
+}
+export const UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchRequest =
+  /*@__PURE__*/ S.suspend(() =>
+    S.Struct({
+      core_tools: S.optional(
+        UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchRequestCoreToolsList,
+      ),
+      embedding_model: S.optional(S.NullOr(S.String)),
+      similarity_threshold: S.optional(S.Number),
+      top_k: S.optional(S.Number),
+    }).pipe(
+      T.Http({
+        method: "PATCH",
+        uri: "/update/mcp_tool_search_settings",
+        code: 200,
+      }),
+    ),
+  ).annotate({
+    identifier:
+      "UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchRequest",
+  }) as any as S.Schema<UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchRequest>;
+
+export type UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchResponseBodyMap =
+  { [key: string]: unknown | undefined };
+export const UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchResponseBodyMap =
+  /*@__PURE__*/ S.Record(
+    S.String,
+    S.Unknown,
+  ) as any as S.Schema<UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchResponseBodyMap>;
+
+export interface UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchResponse {
+  body: UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchResponseBodyMap;
+}
+export const UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchResponse =
+  /*@__PURE__*/ S.suspend(() =>
+    S.Struct({
+      body: UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchResponseBodyMap,
+    }),
+  ).annotate({
+    identifier:
+      "UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchResponse",
+  }) as any as S.Schema<UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchResponse>;
+
+/** LLM providers to intercept for (e.g. 'bedrock', 'vertex_ai'). Empty intercepts Bedrock only. */
+export type UpdateWebsearchInterceptionSettingsUpdateWebsearchInterceptionSettingsPatchRequestEnabledProvidersList =
+  Array<string>;
+export const UpdateWebsearchInterceptionSettingsUpdateWebsearchInterceptionSettingsPatchRequestEnabledProvidersList =
+  /*@__PURE__*/ S.Array(
+    S.String,
+  ) as any as S.Schema<UpdateWebsearchInterceptionSettingsUpdateWebsearchInterceptionSettingsPatchRequestEnabledProvidersList>;
+
+export interface UpdateWebsearchInterceptionSettingsUpdateWebsearchInterceptionSettingsPatchRequest {
+  /** Serve web search tool calls from a configured search tool instead of passing them upstream */
+  enabled?: boolean;
+  /** LLM providers to intercept for (e.g. 'bedrock', 'vertex_ai'). Empty intercepts Bedrock only. */
+  enabled_providers?: UpdateWebsearchInterceptionSettingsUpdateWebsearchInterceptionSettingsPatchRequestEnabledProvidersList;
+  /** How many follow-up model calls one intercepted request may chain. Empty applies the default of 3. */
+  max_agentic_loops?: number | null;
+  /** Name of the configured search tool to run searches through. Empty uses the first one available. */
+  search_tool_name?: string | null;
+}
+export const UpdateWebsearchInterceptionSettingsUpdateWebsearchInterceptionSettingsPatchRequest =
+  /*@__PURE__*/ S.suspend(() =>
+    S.Struct({
+      enabled: S.optional(S.Boolean),
+      enabled_providers: S.optional(
+        UpdateWebsearchInterceptionSettingsUpdateWebsearchInterceptionSettingsPatchRequestEnabledProvidersList,
+      ),
+      max_agentic_loops: S.optional(S.NullOr(S.Number)),
+      search_tool_name: S.optional(S.NullOr(S.String)),
+    }).pipe(
+      T.Http({
+        method: "PATCH",
+        uri: "/update/websearch_interception_settings",
+        code: 200,
+      }),
+    ),
+  ).annotate({
+    identifier:
+      "UpdateWebsearchInterceptionSettingsUpdateWebsearchInterceptionSettingsPatchRequest",
+  }) as any as S.Schema<UpdateWebsearchInterceptionSettingsUpdateWebsearchInterceptionSettingsPatchRequest>;
+
+export interface UpdateWebsearchInterceptionSettingsUpdateWebsearchInterceptionSettingsPatchResponse {
+  body: unknown;
+}
+export const UpdateWebsearchInterceptionSettingsUpdateWebsearchInterceptionSettingsPatchResponse =
+  /*@__PURE__*/ S.suspend(() =>
+    S.Struct({
+      body: S.Unknown,
+    }),
+  ).annotate({
+    identifier:
+      "UpdateWebsearchInterceptionSettingsUpdateWebsearchInterceptionSettingsPatchResponse",
+  }) as any as S.Schema<UpdateWebsearchInterceptionSettingsUpdateWebsearchInterceptionSettingsPatchResponse>;
+
 export type GetMcpSemanticFilterSettingsGetMcpSemanticFilterSettingsGetError =
   LitellmOpError;
 /** Get Mcp Semantic Filter Settings Get MCP semantic filter configuration. Returns current settings for semantic tool filtering. */
@@ -126,6 +334,39 @@ export const getMcpSemanticFilterSettingsGetMcpSemanticFilterSettingsGet: API.Op
   retry: Retry.Retry,
 }));
 
+export type GetMcpToolSearchSettingsGetMcpToolSearchSettingsGetError =
+  LitellmOpError;
+/** Get Mcp Tool Search Settings Get the `litellm_settings.mcp_tool_search` configuration used by the native `mcp_tool_search` virtual tool. */
+export const getMcpToolSearchSettingsGetMcpToolSearchSettingsGet: API.OperationMethod<
+  GetMcpToolSearchSettingsGetMcpToolSearchSettingsGetRequest,
+  MCPToolSearchSettingsResponse,
+  GetMcpToolSearchSettingsGetMcpToolSearchSettingsGetError,
+  LitellmOpContext
+> = /*@__PURE__*/ API.make(() => ({
+  input: GetMcpToolSearchSettingsGetMcpToolSearchSettingsGetRequest,
+  output: MCPToolSearchSettingsResponse,
+  errors: [],
+  protocol: LitellmProtocol,
+  retry: Retry.Retry,
+}));
+
+export type GetWebsearchInterceptionSettingsGetWebsearchInterceptionSettingsGetError =
+  LitellmOpError;
+/** Get Websearch Interception Settings Get web search interception configuration. Returns the current settings plus their schema, for the Admin UI to render. */
+export const getWebsearchInterceptionSettingsGetWebsearchInterceptionSettingsGet: API.OperationMethod<
+  GetWebsearchInterceptionSettingsGetWebsearchInterceptionSettingsGetRequest,
+  WebSearchInterceptionSettingsResponse,
+  GetWebsearchInterceptionSettingsGetWebsearchInterceptionSettingsGetError,
+  LitellmOpContext
+> = /*@__PURE__*/ API.make(() => ({
+  input:
+    GetWebsearchInterceptionSettingsGetWebsearchInterceptionSettingsGetRequest,
+  output: WebSearchInterceptionSettingsResponse,
+  errors: [],
+  protocol: LitellmProtocol,
+  retry: Retry.Retry,
+}));
+
 export type UpdateMcpSemanticFilterSettingsUpdateMcpSemanticFilterSettingsPatchError =
   | UnprocessableEntity
   | LitellmOpError;
@@ -144,3 +385,39 @@ export const updateMcpSemanticFilterSettingsUpdateMcpSemanticFilterSettingsPatch
   protocol: LitellmProtocol,
   retry: Retry.Retry,
 }));
+
+export type UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchError =
+  | UnprocessableEntity
+  | LitellmOpError;
+/** Update Mcp Tool Search Settings Update `litellm_settings.mcp_tool_search` in the database. Settings will be picked up by all pods within approximately 10 seconds via background polling. */
+export const updateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatch: API.OperationMethod<
+  UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchRequest,
+  UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchResponse,
+  UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchError,
+  LitellmOpContext
+> = /*@__PURE__*/ API.make(() => ({
+  input: UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchRequest,
+  output: UpdateMcpToolSearchSettingsUpdateMcpToolSearchSettingsPatchResponse,
+  errors: [UnprocessableEntity],
+  protocol: LitellmProtocol,
+  retry: Retry.Retry,
+}));
+
+export type UpdateWebsearchInterceptionSettingsUpdateWebsearchInterceptionSettingsPatchError =
+  | UnprocessableEntity
+  | LitellmOpError;
+/** Update Websearch Interception Settings Update web search interception settings in database. Settings will be picked up by all pods within approximately 10 seconds via background polling. */
+export const updateWebsearchInterceptionSettingsUpdateWebsearchInterceptionSettingsPatch: API.OperationMethod<
+  UpdateWebsearchInterceptionSettingsUpdateWebsearchInterceptionSettingsPatchRequest,
+  UpdateWebsearchInterceptionSettingsUpdateWebsearchInterceptionSettingsPatchResponse,
+  UpdateWebsearchInterceptionSettingsUpdateWebsearchInterceptionSettingsPatchError,
+  LitellmOpContext
+> = /*@__PURE__*/ API.make(() => ({
+  input:
+    UpdateWebsearchInterceptionSettingsUpdateWebsearchInterceptionSettingsPatchRequest,
+  output:
+    UpdateWebsearchInterceptionSettingsUpdateWebsearchInterceptionSettingsPatchResponse,
+  errors: [UnprocessableEntity],
+  protocol: LitellmProtocol,
+  retry: Retry.Retry,
+}));

package/src/services/sso_settings.ts

@@ -1,4 +1,4 @@
-// AUTO-GENERATED by scripts/generate.ts from .generated-specs (pinned litellm-openapi @ 1.100.0 — see README). Do not edit.
+// AUTO-GENERATED by scripts/generate.ts from .generated-specs (pinned litellm-openapi @ 1.103.0 — see README). Do not edit.
 import * as S from "@distilled.cloud/core/schema";
 import * as API from "@distilled.cloud/core/api";
 import * as C from "@distilled.cloud/core/category";
@@ -174,6 +174,7 @@ export type KeyManagementRoutes =
   | "/team/key/bulk_update"
   | "/key/{key_id}/reset_spend"
   | "/key/access_group_assignment"
+  | "/auto_router/manage"
   | "/key/info"
   | "/key/health"
   | "/key/list"

package/src/services/tag_management.ts

@@ -1,4 +1,4 @@
-// AUTO-GENERATED by scripts/generate.ts from .generated-specs (pinned litellm-openapi @ 1.100.0 — see README). Do not edit.
+// AUTO-GENERATED by scripts/generate.ts from .generated-specs (pinned litellm-openapi @ 1.103.0 — see README). Do not edit.
 import * as S from "@distilled.cloud/core/schema";
 import * as API from "@distilled.cloud/core/api";
 import * as C from "@distilled.cloud/core/category";
@@ -295,8 +295,10 @@ export interface DailySpendMetadata {
   total_pages?: number;
   total_prompt_caching_savings_spend?: number;
   total_prompt_tokens?: number;
+  total_response_time_ms?: number;
   total_spend?: number;
   total_successful_requests?: number;
+  total_timed_requests?: number;
   total_tokens?: number;
 }
 export const DailySpendMetadata = /*@__PURE__*/ S.suspend(() =>
@@ -316,8 +318,10 @@ export const DailySpendMetadata = /*@__PURE__*/ S.suspend(() =>
     total_pages: S.optional(S.Number),
     total_prompt_caching_savings_spend: S.optional(S.Number),
     total_prompt_tokens: S.optional(S.Number),
+    total_response_time_ms: S.optional(S.Number),
     total_spend: S.optional(S.Number),
     total_successful_requests: S.optional(S.Number),
+    total_timed_requests: S.optional(S.Number),
     total_tokens: S.optional(S.Number),
   }),
 ).annotate({
@@ -327,12 +331,18 @@ export const DailySpendMetadata = /*@__PURE__*/ S.suspend(() =>
 /** Metadata for a key */
 export interface KeyMetadata {
   key_alias?: string | null;
+  key_exists?: boolean | null;
   team_id?: string | null;
+  user_email?: string | null;
+  user_id?: string | null;
 }
 export const KeyMetadata = /*@__PURE__*/ S.suspend(() =>
   S.Struct({
     key_alias: S.optional(S.NullOr(S.String)),
+    key_exists: S.optional(S.NullOr(S.Boolean)),
     team_id: S.optional(S.NullOr(S.String)),
+    user_email: S.optional(S.NullOr(S.String)),
+    user_id: S.optional(S.NullOr(S.String)),
   }),
 ).annotate({ identifier: "KeyMetadata" }) as any as S.Schema<KeyMetadata>;
 
@@ -351,6 +361,8 @@ export interface SpendMetrics {
   prompt_tokens?: number;
   spend?: number;
   successful_requests?: number;
+  timed_requests?: number;
+  total_response_time_ms?: number;
   total_tokens?: number;
 }
 export const SpendMetrics = /*@__PURE__*/ S.suspend(() =>
@@ -369,6 +381,8 @@ export const SpendMetrics = /*@__PURE__*/ S.suspend(() =>
     prompt_tokens: S.optional(S.Number),
     spend: S.optional(S.Number),
     successful_requests: S.optional(S.Number),
+    timed_requests: S.optional(S.Number),
+    total_response_time_ms: S.optional(S.Number),
     total_tokens: S.optional(S.Number),
   }),
 ).annotate({ identifier: "SpendMetrics" }) as any as S.Schema<SpendMetrics>;