car-runtime 0.16.1 → 0.17.0

package/index.d.ts

@@ -242,10 +242,19 @@ export class CarRuntime {
    *
    * **Note:** intent is not exposed on the tracked path until the
    * positional argument list is converted to an options object —
-   * this method already takes 8 positional parameters and adding a
-   * 9th would push call sites past readability. For new code, use
-   * {@link inferTrackedWithRequest} which takes a JSON-stringified
-   * `GenerateRequest` and exposes every field including `intent`.
+   * this method already takes 9 positional parameters and adding
+   * intent would push call sites past readability. For new code,
+   * use {@link inferTrackedWithRequest} which takes a JSON-
+   * stringified `GenerateRequest` and exposes every field
+   * including `intent`.
+   *
+   * `imagesJson` is a JSON-encoded array of `ContentBlock` image
+   * variants — either
+   * `{ "type": "image_base64", "data": "<b64>", "media_type": "image/png" }`
+   * or `{ "type": "image_url", "url": "https://…", "detail": "auto" }`.
+   * Vision-capable hosted models (Claude 3.5+/4.x, GPT-4o, Gemini)
+   * accept these directly; non-vision providers reject the request
+   * via a structured error from the daemon. See #230.
    */
   inferTracked(
     prompt: string,
@@ -256,6 +265,7 @@ export class CarRuntime {
     messagesJson?: string | null,
     toolChoice?: string | null,
     parallelToolCalls?: boolean | null,
+    imagesJson?: string | null,
   ): Promise<string>;
 
   /**
@@ -895,6 +905,26 @@ export function transcribeStreamPush(
 
 export function listVoiceSessions(rt: CarRuntime): string;
 
+/**
+ * Start a streaming TTS synthesis.
+ *
+ * Stub: not exposed in the FFI bindings. Connect to the daemon's
+ * WebSocket and use `voice.tts_stream.start`; chunks arrive as
+ * `voice.event` notifications with `type = "tts_chunk"`.
+ */
+export function ttsStreamStart(
+  rt: CarRuntime,
+  streamId: string,
+  text: string,
+  optionsJson?: string | null,
+): Promise<string>;
+
+/** Cancel an in-flight TTS stream. Idempotent. */
+export function ttsStreamCancel(rt: CarRuntime, streamId: string): Promise<string>;
+
+/** List the ids of all in-flight TTS streams. */
+export function listTtsStreams(rt: CarRuntime): string;
+
 // --- Voice turn dispatch (two-track sidecar pattern) ---
 
 export interface DispatchVoiceTurnRequest {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "car-runtime",
-  "version": "0.16.1",
+  "version": "0.17.0",
   "description": "Common Agent Runtime — a deterministic execution layer for AI agents",
   "main": "index.js",
   "types": "index.d.ts",