@homeflare/distilled-litellm 0.2.0

package/src/services/rag.ts

@@ -0,0 +1,148 @@
+// AUTO-GENERATED by scripts/generate.ts from .generated-specs (pinned litellm-openapi @ 1.100.0 — see README). Do not edit.
+import * as S from "@distilled.cloud/core/schema";
+import * as API from "@distilled.cloud/core/api";
+import * as T from "../traits.ts";
+import {
+  LitellmProtocol,
+  type LitellmOpError,
+  type LitellmOpContext,
+} from "../protocol.ts";
+import * as Retry from "../retry.ts";
+
+export type { LitellmOpError, LitellmOpContext };
+
+export interface PostRagIngestRagIngestRequest {}
+export const PostRagIngestRagIngestRequest = /*@__PURE__*/ S.suspend(() =>
+  S.Struct({}).pipe(T.Http({ method: "POST", uri: "/rag/ingest", code: 200 })),
+).annotate({
+  identifier: "PostRagIngestRagIngestRequest",
+}) as any as S.Schema<PostRagIngestRagIngestRequest>;
+
+export interface PostRagIngestRagIngestResponse {
+  body: unknown;
+}
+export const PostRagIngestRagIngestResponse = /*@__PURE__*/ S.suspend(() =>
+  S.Struct({
+    body: S.Unknown,
+  }),
+).annotate({
+  identifier: "PostRagIngestRagIngestResponse",
+}) as any as S.Schema<PostRagIngestRagIngestResponse>;
+
+export interface PostRagIngestV1RagIngestRequest {}
+export const PostRagIngestV1RagIngestRequest = /*@__PURE__*/ S.suspend(() =>
+  S.Struct({}).pipe(
+    T.Http({ method: "POST", uri: "/v1/rag/ingest", code: 200 }),
+  ),
+).annotate({
+  identifier: "PostRagIngestV1RagIngestRequest",
+}) as any as S.Schema<PostRagIngestV1RagIngestRequest>;
+
+export interface PostRagIngestV1RagIngestResponse {
+  body: unknown;
+}
+export const PostRagIngestV1RagIngestResponse = /*@__PURE__*/ S.suspend(() =>
+  S.Struct({
+    body: S.Unknown,
+  }),
+).annotate({
+  identifier: "PostRagIngestV1RagIngestResponse",
+}) as any as S.Schema<PostRagIngestV1RagIngestResponse>;
+
+export interface PostRagQueryRagQueryRequest {}
+export const PostRagQueryRagQueryRequest = /*@__PURE__*/ S.suspend(() =>
+  S.Struct({}).pipe(T.Http({ method: "POST", uri: "/rag/query", code: 200 })),
+).annotate({
+  identifier: "PostRagQueryRagQueryRequest",
+}) as any as S.Schema<PostRagQueryRagQueryRequest>;
+
+export interface PostRagQueryRagQueryResponse {
+  body: unknown;
+}
+export const PostRagQueryRagQueryResponse = /*@__PURE__*/ S.suspend(() =>
+  S.Struct({
+    body: S.Unknown,
+  }),
+).annotate({
+  identifier: "PostRagQueryRagQueryResponse",
+}) as any as S.Schema<PostRagQueryRagQueryResponse>;
+
+export interface PostRagQueryV1RagQueryRequest {}
+export const PostRagQueryV1RagQueryRequest = /*@__PURE__*/ S.suspend(() =>
+  S.Struct({}).pipe(
+    T.Http({ method: "POST", uri: "/v1/rag/query", code: 200 }),
+  ),
+).annotate({
+  identifier: "PostRagQueryV1RagQueryRequest",
+}) as any as S.Schema<PostRagQueryV1RagQueryRequest>;
+
+export interface PostRagQueryV1RagQueryResponse {
+  body: unknown;
+}
+export const PostRagQueryV1RagQueryResponse = /*@__PURE__*/ S.suspend(() =>
+  S.Struct({
+    body: S.Unknown,
+  }),
+).annotate({
+  identifier: "PostRagQueryV1RagQueryResponse",
+}) as any as S.Schema<PostRagQueryV1RagQueryResponse>;
+
+export type PostRagIngestRagIngestError = LitellmOpError;
+/** Rag Ingest RAG Ingest endpoint - all-in-one document ingestion pipeline. Supports form upload (for files) or JSON body (for URLs). ## Form upload (for files): ```bash curl -X POST "http://localhost:4000/v1/rag/ingest" \ -H "Authorization: Bearer sk-1234" \ -F file="@document.pdf" \ -F 'ingest_options={"vector_store": {"custom_llm_provider": "openai"}}' ``` ## JSON body (for URLs): ```bash curl -X POST "http://localhost:4000/v1/rag/ingest" \ -H "Authorization: Bearer sk-1234" \ -H "Content-Type: application/json" \ -d '{ "file_url": "https://example.com/document.pdf", "ingest_options": {"vector_store": {"custom_llm_provider": "openai"}} }' ``` ## Bedrock: ```bash curl -X POST "http://localhost:4000/v1/rag/ingest" \ -H "Authorization: Bearer sk-1234" \ -F file="@document.pdf" \ -F 'ingest_options={"vector_store": {"custom_llm_provider": "bedrock"}}' ``` */
+export const postRagIngestRagIngest: API.OperationMethod<
+  PostRagIngestRagIngestRequest,
+  PostRagIngestRagIngestResponse,
+  PostRagIngestRagIngestError,
+  LitellmOpContext
+> = /*@__PURE__*/ API.make(() => ({
+  input: PostRagIngestRagIngestRequest,
+  output: PostRagIngestRagIngestResponse,
+  errors: [],
+  protocol: LitellmProtocol,
+  retry: Retry.Retry,
+}));
+
+export type PostRagIngestV1RagIngestError = LitellmOpError;
+/** Rag Ingest RAG Ingest endpoint - all-in-one document ingestion pipeline. Supports form upload (for files) or JSON body (for URLs). ## Form upload (for files): ```bash curl -X POST "http://localhost:4000/v1/rag/ingest" \ -H "Authorization: Bearer sk-1234" \ -F file="@document.pdf" \ -F 'ingest_options={"vector_store": {"custom_llm_provider": "openai"}}' ``` ## JSON body (for URLs): ```bash curl -X POST "http://localhost:4000/v1/rag/ingest" \ -H "Authorization: Bearer sk-1234" \ -H "Content-Type: application/json" \ -d '{ "file_url": "https://example.com/document.pdf", "ingest_options": {"vector_store": {"custom_llm_provider": "openai"}} }' ``` ## Bedrock: ```bash curl -X POST "http://localhost:4000/v1/rag/ingest" \ -H "Authorization: Bearer sk-1234" \ -F file="@document.pdf" \ -F 'ingest_options={"vector_store": {"custom_llm_provider": "bedrock"}}' ``` */
+export const postRagIngestV1RagIngest: API.OperationMethod<
+  PostRagIngestV1RagIngestRequest,
+  PostRagIngestV1RagIngestResponse,
+  PostRagIngestV1RagIngestError,
+  LitellmOpContext
+> = /*@__PURE__*/ API.make(() => ({
+  input: PostRagIngestV1RagIngestRequest,
+  output: PostRagIngestV1RagIngestResponse,
+  errors: [],
+  protocol: LitellmProtocol,
+  retry: Retry.Retry,
+}));
+
+export type PostRagQueryRagQueryError = LitellmOpError;
+/** Rag Query RAG Query endpoint - search vector store, optionally rerank, and generate LLM response. This endpoint: 1. Extracts the query from the last user message 2. Searches the vector store for relevant context 3. Optionally reranks the results 4. Generates an LLM response with the retrieved context ## Example Request: ```bash curl -X POST "http://localhost:4000/v1/rag/query" \ -H "Authorization: Bearer sk-1234" \ -H "Content-Type: application/json" \ -d '{ "model": "gpt-4o-mini", "messages": [{"role": "user", "content": "What is LiteLLM?"}], "retrieval_config": { "vector_store_id": "vs_abc123", "custom_llm_provider": "openai", "top_k": 5 } }' ``` ## With Reranking: ```bash curl -X POST "http://localhost:4000/v1/rag/query" \ -H "Authorization: Bearer sk-1234" \ -H "Content-Type: application/json" \ -d '{ "model": "gpt-4o-mini", "messages": [{"role": "user", "content": "What is LiteLLM?"}], "retrieval_config": { "vector_store_id": "vs_abc123", "custom_llm_provider": "openai", "top_k": 10 }, "rerank": { "enabled": true, "model": "cohere/rerank-english-v3.0", "top_n": 3 } }' ``` */
+export const postRagQueryRagQuery: API.OperationMethod<
+  PostRagQueryRagQueryRequest,
+  PostRagQueryRagQueryResponse,
+  PostRagQueryRagQueryError,
+  LitellmOpContext
+> = /*@__PURE__*/ API.make(() => ({
+  input: PostRagQueryRagQueryRequest,
+  output: PostRagQueryRagQueryResponse,
+  errors: [],
+  protocol: LitellmProtocol,
+  retry: Retry.Retry,
+}));
+
+export type PostRagQueryV1RagQueryError = LitellmOpError;
+/** Rag Query RAG Query endpoint - search vector store, optionally rerank, and generate LLM response. This endpoint: 1. Extracts the query from the last user message 2. Searches the vector store for relevant context 3. Optionally reranks the results 4. Generates an LLM response with the retrieved context ## Example Request: ```bash curl -X POST "http://localhost:4000/v1/rag/query" \ -H "Authorization: Bearer sk-1234" \ -H "Content-Type: application/json" \ -d '{ "model": "gpt-4o-mini", "messages": [{"role": "user", "content": "What is LiteLLM?"}], "retrieval_config": { "vector_store_id": "vs_abc123", "custom_llm_provider": "openai", "top_k": 5 } }' ``` ## With Reranking: ```bash curl -X POST "http://localhost:4000/v1/rag/query" \ -H "Authorization: Bearer sk-1234" \ -H "Content-Type: application/json" \ -d '{ "model": "gpt-4o-mini", "messages": [{"role": "user", "content": "What is LiteLLM?"}], "retrieval_config": { "vector_store_id": "vs_abc123", "custom_llm_provider": "openai", "top_k": 10 }, "rerank": { "enabled": true, "model": "cohere/rerank-english-v3.0", "top_n": 3 } }' ``` */
+export const postRagQueryV1RagQuery: API.OperationMethod<
+  PostRagQueryV1RagQueryRequest,
+  PostRagQueryV1RagQueryResponse,
+  PostRagQueryV1RagQueryError,
+  LitellmOpContext
+> = /*@__PURE__*/ API.make(() => ({
+  input: PostRagQueryV1RagQueryRequest,
+  output: PostRagQueryV1RagQueryResponse,
+  errors: [],
+  protocol: LitellmProtocol,
+  retry: Retry.Retry,
+}));

package/src/services/realtime.ts

@@ -0,0 +1,350 @@
+// AUTO-GENERATED by scripts/generate.ts from .generated-specs (pinned litellm-openapi @ 1.100.0 — see README). Do not edit.
+import * as S from "@distilled.cloud/core/schema";
+import * as API from "@distilled.cloud/core/api";
+import * as T from "../traits.ts";
+import {
+  LitellmProtocol,
+  type LitellmOpError,
+  type LitellmOpContext,
+} from "../protocol.ts";
+import * as Retry from "../retry.ts";
+
+export type { LitellmOpError, LitellmOpContext };
+
+export interface CreateRealtimeClientSecretOpenaiV1RealtimeClientSecretsPostRequest {}
+export const CreateRealtimeClientSecretOpenaiV1RealtimeClientSecretsPostRequest =
+  /*@__PURE__*/ S.suspend(() =>
+    S.Struct({}).pipe(
+      T.Http({
+        method: "POST",
+        uri: "/openai/v1/realtime/client_secrets",
+        code: 200,
+      }),
+    ),
+  ).annotate({
+    identifier:
+      "CreateRealtimeClientSecretOpenaiV1RealtimeClientSecretsPostRequest",
+  }) as any as S.Schema<CreateRealtimeClientSecretOpenaiV1RealtimeClientSecretsPostRequest>;
+
+export type RealtimeClientSecretResponseSessionMap = {
+  [key: string]: unknown | undefined;
+};
+export const RealtimeClientSecretResponseSessionMap = /*@__PURE__*/ S.Record(
+  S.String,
+  S.Unknown,
+) as any as S.Schema<RealtimeClientSecretResponseSessionMap>;
+
+/** Response from POST /v1/realtime/client_secrets. Both the top-level `value` and `session.client_secret.value` will contain the encrypted token instead of the raw ephemeral key. The `session` field is kept as a raw dict so unknown fields pass through. */
+export interface RealtimeClientSecretResponse {
+  expires_at?: number | null;
+  session?: RealtimeClientSecretResponseSessionMap | null;
+  value: string;
+}
+export const RealtimeClientSecretResponse = /*@__PURE__*/ S.suspend(() =>
+  S.Struct({
+    expires_at: S.optional(S.NullOr(S.Number)),
+    session: S.optional(S.NullOr(RealtimeClientSecretResponseSessionMap)),
+    value: S.String,
+  }),
+).annotate({
+  identifier: "RealtimeClientSecretResponse",
+}) as any as S.Schema<RealtimeClientSecretResponse>;
+
+export interface CreateRealtimeClientSecretRealtimeClientSecretsPostRequest {}
+export const CreateRealtimeClientSecretRealtimeClientSecretsPostRequest =
+  /*@__PURE__*/ S.suspend(() =>
+    S.Struct({}).pipe(
+      T.Http({ method: "POST", uri: "/realtime/client_secrets", code: 200 }),
+    ),
+  ).annotate({
+    identifier: "CreateRealtimeClientSecretRealtimeClientSecretsPostRequest",
+  }) as any as S.Schema<CreateRealtimeClientSecretRealtimeClientSecretsPostRequest>;
+
+export interface CreateRealtimeClientSecretV1RealtimeClientSecretsPostRequest {}
+export const CreateRealtimeClientSecretV1RealtimeClientSecretsPostRequest =
+  /*@__PURE__*/ S.suspend(() =>
+    S.Struct({}).pipe(
+      T.Http({ method: "POST", uri: "/v1/realtime/client_secrets", code: 200 }),
+    ),
+  ).annotate({
+    identifier: "CreateRealtimeClientSecretV1RealtimeClientSecretsPostRequest",
+  }) as any as S.Schema<CreateRealtimeClientSecretV1RealtimeClientSecretsPostRequest>;
+
+export interface CreateRealtimeTranscriptionSessionOpenaiV1RealtimeTranscriptionSessionsPostRequest {}
+export const CreateRealtimeTranscriptionSessionOpenaiV1RealtimeTranscriptionSessionsPostRequest =
+  /*@__PURE__*/ S.suspend(() =>
+    S.Struct({}).pipe(
+      T.Http({
+        method: "POST",
+        uri: "/openai/v1/realtime/transcription_sessions",
+        code: 200,
+      }),
+    ),
+  ).annotate({
+    identifier:
+      "CreateRealtimeTranscriptionSessionOpenaiV1RealtimeTranscriptionSessionsPostRequest",
+  }) as any as S.Schema<CreateRealtimeTranscriptionSessionOpenaiV1RealtimeTranscriptionSessionsPostRequest>;
+
+export type RealtimeTranscriptionSessionResponseClientSecretMap = {
+  [key: string]: unknown | undefined;
+};
+export const RealtimeTranscriptionSessionResponseClientSecretMap =
+  /*@__PURE__*/ S.Record(
+    S.String,
+    S.Unknown,
+  ) as any as S.Schema<RealtimeTranscriptionSessionResponseClientSecretMap>;
+
+/** Response from POST /v1/realtime/transcription_sessions. `client_secret.value` contains the encrypted token instead of the raw ephemeral key. Unknown fields pass through unchanged. */
+export interface RealtimeTranscriptionSessionResponse {
+  client_secret?: RealtimeTranscriptionSessionResponseClientSecretMap | null;
+}
+export const RealtimeTranscriptionSessionResponse = /*@__PURE__*/ S.suspend(
+  () =>
+    S.Struct({
+      client_secret: S.optional(
+        S.NullOr(RealtimeTranscriptionSessionResponseClientSecretMap),
+      ),
+    }),
+).annotate({
+  identifier: "RealtimeTranscriptionSessionResponse",
+}) as any as S.Schema<RealtimeTranscriptionSessionResponse>;
+
+export interface CreateRealtimeTranscriptionSessionRealtimeTranscriptionSessionsPostRequest {}
+export const CreateRealtimeTranscriptionSessionRealtimeTranscriptionSessionsPostRequest =
+  /*@__PURE__*/ S.suspend(() =>
+    S.Struct({}).pipe(
+      T.Http({
+        method: "POST",
+        uri: "/realtime/transcription_sessions",
+        code: 200,
+      }),
+    ),
+  ).annotate({
+    identifier:
+      "CreateRealtimeTranscriptionSessionRealtimeTranscriptionSessionsPostRequest",
+  }) as any as S.Schema<CreateRealtimeTranscriptionSessionRealtimeTranscriptionSessionsPostRequest>;
+
+export interface CreateRealtimeTranscriptionSessionV1RealtimeTranscriptionSessionsPostRequest {}
+export const CreateRealtimeTranscriptionSessionV1RealtimeTranscriptionSessionsPostRequest =
+  /*@__PURE__*/ S.suspend(() =>
+    S.Struct({}).pipe(
+      T.Http({
+        method: "POST",
+        uri: "/v1/realtime/transcription_sessions",
+        code: 200,
+      }),
+    ),
+  ).annotate({
+    identifier:
+      "CreateRealtimeTranscriptionSessionV1RealtimeTranscriptionSessionsPostRequest",
+  }) as any as S.Schema<CreateRealtimeTranscriptionSessionV1RealtimeTranscriptionSessionsPostRequest>;
+
+export interface PostProxyRealtimeCallsOpenaiV1RealtimeCallRequest {}
+export const PostProxyRealtimeCallsOpenaiV1RealtimeCallRequest =
+  /*@__PURE__*/ S.suspend(() =>
+    S.Struct({}).pipe(
+      T.Http({ method: "POST", uri: "/openai/v1/realtime/calls", code: 200 }),
+    ),
+  ).annotate({
+    identifier: "PostProxyRealtimeCallsOpenaiV1RealtimeCallRequest",
+  }) as any as S.Schema<PostProxyRealtimeCallsOpenaiV1RealtimeCallRequest>;
+
+export interface PostProxyRealtimeCallsOpenaiV1RealtimeCallResponse {
+  body: unknown;
+}
+export const PostProxyRealtimeCallsOpenaiV1RealtimeCallResponse =
+  /*@__PURE__*/ S.suspend(() =>
+    S.Struct({
+      body: S.Unknown,
+    }),
+  ).annotate({
+    identifier: "PostProxyRealtimeCallsOpenaiV1RealtimeCallResponse",
+  }) as any as S.Schema<PostProxyRealtimeCallsOpenaiV1RealtimeCallResponse>;
+
+export interface PostProxyRealtimeCallsRealtimeCallRequest {}
+export const PostProxyRealtimeCallsRealtimeCallRequest =
+  /*@__PURE__*/ S.suspend(() =>
+    S.Struct({}).pipe(
+      T.Http({ method: "POST", uri: "/realtime/calls", code: 200 }),
+    ),
+  ).annotate({
+    identifier: "PostProxyRealtimeCallsRealtimeCallRequest",
+  }) as any as S.Schema<PostProxyRealtimeCallsRealtimeCallRequest>;
+
+export interface PostProxyRealtimeCallsRealtimeCallResponse {
+  body: unknown;
+}
+export const PostProxyRealtimeCallsRealtimeCallResponse =
+  /*@__PURE__*/ S.suspend(() =>
+    S.Struct({
+      body: S.Unknown,
+    }),
+  ).annotate({
+    identifier: "PostProxyRealtimeCallsRealtimeCallResponse",
+  }) as any as S.Schema<PostProxyRealtimeCallsRealtimeCallResponse>;
+
+export interface PostProxyRealtimeCallsV1RealtimeCallRequest {}
+export const PostProxyRealtimeCallsV1RealtimeCallRequest =
+  /*@__PURE__*/ S.suspend(() =>
+    S.Struct({}).pipe(
+      T.Http({ method: "POST", uri: "/v1/realtime/calls", code: 200 }),
+    ),
+  ).annotate({
+    identifier: "PostProxyRealtimeCallsV1RealtimeCallRequest",
+  }) as any as S.Schema<PostProxyRealtimeCallsV1RealtimeCallRequest>;
+
+export interface PostProxyRealtimeCallsV1RealtimeCallResponse {
+  body: unknown;
+}
+export const PostProxyRealtimeCallsV1RealtimeCallResponse =
+  /*@__PURE__*/ S.suspend(() =>
+    S.Struct({
+      body: S.Unknown,
+    }),
+  ).annotate({
+    identifier: "PostProxyRealtimeCallsV1RealtimeCallResponse",
+  }) as any as S.Schema<PostProxyRealtimeCallsV1RealtimeCallResponse>;
+
+export type CreateRealtimeClientSecretOpenaiV1RealtimeClientSecretsPostError =
+  LitellmOpError;
+/** Create Realtime Client Secret */
+export const createRealtimeClientSecretOpenaiV1RealtimeClientSecretsPost: API.OperationMethod<
+  CreateRealtimeClientSecretOpenaiV1RealtimeClientSecretsPostRequest,
+  RealtimeClientSecretResponse,
+  CreateRealtimeClientSecretOpenaiV1RealtimeClientSecretsPostError,
+  LitellmOpContext
+> = /*@__PURE__*/ API.make(() => ({
+  input: CreateRealtimeClientSecretOpenaiV1RealtimeClientSecretsPostRequest,
+  output: RealtimeClientSecretResponse,
+  errors: [],
+  protocol: LitellmProtocol,
+  retry: Retry.Retry,
+}));
+
+export type CreateRealtimeClientSecretRealtimeClientSecretsPostError =
+  LitellmOpError;
+/** Create Realtime Client Secret */
+export const createRealtimeClientSecretRealtimeClientSecretsPost: API.OperationMethod<
+  CreateRealtimeClientSecretRealtimeClientSecretsPostRequest,
+  RealtimeClientSecretResponse,
+  CreateRealtimeClientSecretRealtimeClientSecretsPostError,
+  LitellmOpContext
+> = /*@__PURE__*/ API.make(() => ({
+  input: CreateRealtimeClientSecretRealtimeClientSecretsPostRequest,
+  output: RealtimeClientSecretResponse,
+  errors: [],
+  protocol: LitellmProtocol,
+  retry: Retry.Retry,
+}));
+
+export type CreateRealtimeClientSecretV1RealtimeClientSecretsPostError =
+  LitellmOpError;
+/** Create Realtime Client Secret */
+export const createRealtimeClientSecretV1RealtimeClientSecretsPost: API.OperationMethod<
+  CreateRealtimeClientSecretV1RealtimeClientSecretsPostRequest,
+  RealtimeClientSecretResponse,
+  CreateRealtimeClientSecretV1RealtimeClientSecretsPostError,
+  LitellmOpContext
+> = /*@__PURE__*/ API.make(() => ({
+  input: CreateRealtimeClientSecretV1RealtimeClientSecretsPostRequest,
+  output: RealtimeClientSecretResponse,
+  errors: [],
+  protocol: LitellmProtocol,
+  retry: Retry.Retry,
+}));
+
+export type CreateRealtimeTranscriptionSessionOpenaiV1RealtimeTranscriptionSessionsPostError =
+  LitellmOpError;
+/** Create Realtime Transcription Session Create an ephemeral Realtime transcription session (POST /v1/realtime/transcription_sessions) for the WebRTC/WebSocket flow. Mirrors the client_secrets route but targets the transcription_sessions endpoint and encrypts the ephemeral key returned under `client_secret.value`. */
+export const createRealtimeTranscriptionSessionOpenaiV1RealtimeTranscriptionSessionsPost: API.OperationMethod<
+  CreateRealtimeTranscriptionSessionOpenaiV1RealtimeTranscriptionSessionsPostRequest,
+  RealtimeTranscriptionSessionResponse,
+  CreateRealtimeTranscriptionSessionOpenaiV1RealtimeTranscriptionSessionsPostError,
+  LitellmOpContext
+> = /*@__PURE__*/ API.make(() => ({
+  input:
+    CreateRealtimeTranscriptionSessionOpenaiV1RealtimeTranscriptionSessionsPostRequest,
+  output: RealtimeTranscriptionSessionResponse,
+  errors: [],
+  protocol: LitellmProtocol,
+  retry: Retry.Retry,
+}));
+
+export type CreateRealtimeTranscriptionSessionRealtimeTranscriptionSessionsPostError =
+  LitellmOpError;
+/** Create Realtime Transcription Session Create an ephemeral Realtime transcription session (POST /v1/realtime/transcription_sessions) for the WebRTC/WebSocket flow. Mirrors the client_secrets route but targets the transcription_sessions endpoint and encrypts the ephemeral key returned under `client_secret.value`. */
+export const createRealtimeTranscriptionSessionRealtimeTranscriptionSessionsPost: API.OperationMethod<
+  CreateRealtimeTranscriptionSessionRealtimeTranscriptionSessionsPostRequest,
+  RealtimeTranscriptionSessionResponse,
+  CreateRealtimeTranscriptionSessionRealtimeTranscriptionSessionsPostError,
+  LitellmOpContext
+> = /*@__PURE__*/ API.make(() => ({
+  input:
+    CreateRealtimeTranscriptionSessionRealtimeTranscriptionSessionsPostRequest,
+  output: RealtimeTranscriptionSessionResponse,
+  errors: [],
+  protocol: LitellmProtocol,
+  retry: Retry.Retry,
+}));
+
+export type CreateRealtimeTranscriptionSessionV1RealtimeTranscriptionSessionsPostError =
+  LitellmOpError;
+/** Create Realtime Transcription Session Create an ephemeral Realtime transcription session (POST /v1/realtime/transcription_sessions) for the WebRTC/WebSocket flow. Mirrors the client_secrets route but targets the transcription_sessions endpoint and encrypts the ephemeral key returned under `client_secret.value`. */
+export const createRealtimeTranscriptionSessionV1RealtimeTranscriptionSessionsPost: API.OperationMethod<
+  CreateRealtimeTranscriptionSessionV1RealtimeTranscriptionSessionsPostRequest,
+  RealtimeTranscriptionSessionResponse,
+  CreateRealtimeTranscriptionSessionV1RealtimeTranscriptionSessionsPostError,
+  LitellmOpContext
+> = /*@__PURE__*/ API.make(() => ({
+  input:
+    CreateRealtimeTranscriptionSessionV1RealtimeTranscriptionSessionsPostRequest,
+  output: RealtimeTranscriptionSessionResponse,
+  errors: [],
+  protocol: LitellmProtocol,
+  retry: Retry.Retry,
+}));
+
+export type PostProxyRealtimeCallsOpenaiV1RealtimeCallError = LitellmOpError;
+/** Proxy Realtime Calls */
+export const postProxyRealtimeCallsOpenaiV1RealtimeCall: API.OperationMethod<
+  PostProxyRealtimeCallsOpenaiV1RealtimeCallRequest,
+  PostProxyRealtimeCallsOpenaiV1RealtimeCallResponse,
+  PostProxyRealtimeCallsOpenaiV1RealtimeCallError,
+  LitellmOpContext
+> = /*@__PURE__*/ API.make(() => ({
+  input: PostProxyRealtimeCallsOpenaiV1RealtimeCallRequest,
+  output: PostProxyRealtimeCallsOpenaiV1RealtimeCallResponse,
+  errors: [],
+  protocol: LitellmProtocol,
+  retry: Retry.Retry,
+}));
+
+export type PostProxyRealtimeCallsRealtimeCallError = LitellmOpError;
+/** Proxy Realtime Calls */
+export const postProxyRealtimeCallsRealtimeCall: API.OperationMethod<
+  PostProxyRealtimeCallsRealtimeCallRequest,
+  PostProxyRealtimeCallsRealtimeCallResponse,
+  PostProxyRealtimeCallsRealtimeCallError,
+  LitellmOpContext
+> = /*@__PURE__*/ API.make(() => ({
+  input: PostProxyRealtimeCallsRealtimeCallRequest,
+  output: PostProxyRealtimeCallsRealtimeCallResponse,
+  errors: [],
+  protocol: LitellmProtocol,
+  retry: Retry.Retry,
+}));
+
+export type PostProxyRealtimeCallsV1RealtimeCallError = LitellmOpError;
+/** Proxy Realtime Calls */
+export const postProxyRealtimeCallsV1RealtimeCall: API.OperationMethod<
+  PostProxyRealtimeCallsV1RealtimeCallRequest,
+  PostProxyRealtimeCallsV1RealtimeCallResponse,
+  PostProxyRealtimeCallsV1RealtimeCallError,
+  LitellmOpContext
+> = /*@__PURE__*/ API.make(() => ({
+  input: PostProxyRealtimeCallsV1RealtimeCallRequest,
+  output: PostProxyRealtimeCallsV1RealtimeCallResponse,
+  errors: [],
+  protocol: LitellmProtocol,
+  retry: Retry.Retry,
+}));

package/src/services/rerank.ts

@@ -0,0 +1,111 @@
+// AUTO-GENERATED by scripts/generate.ts from .generated-specs (pinned litellm-openapi @ 1.100.0 — see README). Do not edit.
+import * as S from "@distilled.cloud/core/schema";
+import * as API from "@distilled.cloud/core/api";
+import * as T from "../traits.ts";
+import {
+  LitellmProtocol,
+  type LitellmOpError,
+  type LitellmOpContext,
+} from "../protocol.ts";
+import * as Retry from "../retry.ts";
+
+export type { LitellmOpError, LitellmOpContext };
+
+export interface PostRerankRerankRequest {}
+export const PostRerankRerankRequest = /*@__PURE__*/ S.suspend(() =>
+  S.Struct({}).pipe(T.Http({ method: "POST", uri: "/rerank", code: 200 })),
+).annotate({
+  identifier: "PostRerankRerankRequest",
+}) as any as S.Schema<PostRerankRerankRequest>;
+
+export interface PostRerankRerankResponse {
+  body: unknown;
+}
+export const PostRerankRerankResponse = /*@__PURE__*/ S.suspend(() =>
+  S.Struct({
+    body: S.Unknown,
+  }),
+).annotate({
+  identifier: "PostRerankRerankResponse",
+}) as any as S.Schema<PostRerankRerankResponse>;
+
+export interface PostRerankV1RerankRequest {}
+export const PostRerankV1RerankRequest = /*@__PURE__*/ S.suspend(() =>
+  S.Struct({}).pipe(T.Http({ method: "POST", uri: "/v1/rerank", code: 200 })),
+).annotate({
+  identifier: "PostRerankV1RerankRequest",
+}) as any as S.Schema<PostRerankV1RerankRequest>;
+
+export interface PostRerankV1RerankResponse {
+  body: unknown;
+}
+export const PostRerankV1RerankResponse = /*@__PURE__*/ S.suspend(() =>
+  S.Struct({
+    body: S.Unknown,
+  }),
+).annotate({
+  identifier: "PostRerankV1RerankResponse",
+}) as any as S.Schema<PostRerankV1RerankResponse>;
+
+export interface PostRerankV2RerankRequest {}
+export const PostRerankV2RerankRequest = /*@__PURE__*/ S.suspend(() =>
+  S.Struct({}).pipe(T.Http({ method: "POST", uri: "/v2/rerank", code: 200 })),
+).annotate({
+  identifier: "PostRerankV2RerankRequest",
+}) as any as S.Schema<PostRerankV2RerankRequest>;
+
+export interface PostRerankV2RerankResponse {
+  body: unknown;
+}
+export const PostRerankV2RerankResponse = /*@__PURE__*/ S.suspend(() =>
+  S.Struct({
+    body: S.Unknown,
+  }),
+).annotate({
+  identifier: "PostRerankV2RerankResponse",
+}) as any as S.Schema<PostRerankV2RerankResponse>;
+
+export type PostRerankRerankError = LitellmOpError;
+/** Rerank */
+export const postRerankRerank: API.OperationMethod<
+  PostRerankRerankRequest,
+  PostRerankRerankResponse,
+  PostRerankRerankError,
+  LitellmOpContext
+> = /*@__PURE__*/ API.make(() => ({
+  input: PostRerankRerankRequest,
+  output: PostRerankRerankResponse,
+  errors: [],
+  protocol: LitellmProtocol,
+  retry: Retry.Retry,
+}));
+
+export type PostRerankV1RerankError = LitellmOpError;
+/** Rerank */
+export const postRerankV1Rerank: API.OperationMethod<
+  PostRerankV1RerankRequest,
+  PostRerankV1RerankResponse,
+  PostRerankV1RerankError,
+  LitellmOpContext
+> = /*@__PURE__*/ API.make(() => ({
+  input: PostRerankV1RerankRequest,
+  output: PostRerankV1RerankResponse,
+  errors: [],
+  protocol: LitellmProtocol,
+  retry: Retry.Retry,
+}));
+
+export type PostRerankV2RerankError = LitellmOpError;
+/** Rerank */
+export const postRerankV2Rerank: API.OperationMethod<
+  PostRerankV2RerankRequest,
+  PostRerankV2RerankResponse,
+  PostRerankV2RerankError,
+  LitellmOpContext
+> = /*@__PURE__*/ API.make(() => ({
+  input: PostRerankV2RerankRequest,
+  output: PostRerankV2RerankResponse,
+  errors: [],
+  protocol: LitellmProtocol,
+  retry: Retry.Retry,
+}));