@punica/editor 1.4.11 → 1.4.13

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@punica/editor",
-  "version": "1.4.11",
+  "version": "1.4.13",
   "description": "Punica Editor",
   "private": false,
   "type": "module",

package/types/punica.module.kernel.ai.d.ts

@@ -200,6 +200,13 @@ declare module 'punica' {
        */
       export interface AgentRunInput {
         query: string;
+        /**
+         * Image inputs for a multimodal (vision) turn, each a `data:` URL
+         * (`data:image/png;base64,…`). Attached to the live user message sent
+         * to the model (see `LlmMessage.images`). Optional and additive; a
+         * text-only run omits it.
+         */
+        images?: string[];
         /**
          * Persistent conversation to continue (`kernel.AI.sessions` id).
          * When set, the run seeds its message history from the session's
@@ -345,6 +352,12 @@ declare module 'punica' {
         id: string;
         role: 'system' | 'user' | 'assistant' | 'tool';
         content: string;
+        /**
+         * Image inputs (`data:` URLs) attached to a user turn, mirroring
+         * `runtime.LlmMessage.images` so a persisted multimodal message can be
+         * replayed into an LLM request directly. Optional and additive.
+         */
+        images?: string[];
         toolCalls?: Array<{
           id: string;
           name: string;

package/types/punica.module.kernel.llm.d.ts

@@ -68,6 +68,12 @@ declare module 'punica' {
         instructionClass: string;
         /** User prompt/query */
         userPrompt: string;
+        /**
+         * Optional image inputs (`data:` URLs) for a multimodal invocation,
+         * attached to the user message (see `runtime.LlmMessage.images`).
+         * Additive; text-only invocations omit it.
+         */
+        images?: string[];
         /** Optional context for the LLM invocation */
         context?: {
           workspaceId?: string;

package/types/punica.module.runtime.llm.d.ts

@@ -14,6 +14,15 @@ declare module 'punica' {
       role: LlmRole;
       content: string;
       name?: string;
+      /**
+       * Image inputs for a multimodal (vision) turn, each a `data:` URL
+       * (`data:image/png;base64,…`). Set on a `role: 'user'` message to send
+       * images alongside `content`. The host transport encodes these to the
+       * vendor wire format (OpenAI content-parts `{ type: 'image_url' }`).
+       * Text-only providers ignore the field. Optional and additive — messages
+       * without images are unchanged on the wire.
+       */
+      images?: string[];
       /**
        * Tool calls requested by an assistant turn (substrate shape). Set when
        * replaying an assistant message that invoked tools back to the model on