@babav/knowledge-core-client 0.22.1 → 0.22.3

package/dist/index.d.ts

@@ -51,6 +51,7 @@ export interface RetrievalContent {
 }
 export interface QueryOverrides {
     generation_model?: string;
+    generation_model_mode?: string;
     max_response_tokens?: number;
     top_k_retrieved_chunks?: number;
     top_k_reranked_chunks?: number;
@@ -68,6 +69,7 @@ export interface QueryOverrides {
 /** Per-request visual overrides (see VisualRequest). */
 export interface VisualOverrides {
     concept_model?: string;
+    concept_model_mode?: string;
     concept_vet_model?: string;
     max_revisions?: number;
     claims_check_model?: string;
@@ -107,6 +109,12 @@ export interface Visual {
         height: number;
     };
 }
+/** Visual stage (streaming): generation has started. MAY be emitted twice — first with
+ *  `pending: null` (started, count unknown), then again with the ACCURATE post-vet count before
+ *  rendering. Take the LATEST value; `visuals_complete.count` is always authoritative. */
+export interface VisualsPending {
+    pending: number | null;
+}
 export interface VisualsMeta {
     count: number;
     dropped: number;
@@ -251,6 +259,7 @@ export interface Agent {
     identity_prompt: string | null;
     response_prompt: string | null;
     generation_model: string | null;
+    generation_model_mode: string | null;
     max_response_tokens: number | null;
     top_k_retrieved_chunks: number | null;
     top_k_reranked_chunks: number | null;
@@ -336,6 +345,10 @@ export interface StreamHandlers {
      *  (native path). Also present in `final`. May be `{}` if scoring was unavailable. */
     onGroundedness?: (g: Groundedness | Record<string, never>) => void;
     onFinal?: (d: QueryResponse) => void;
+    /** First event of the trailing visual stage — visual generation has STARTED (fires after
+     *  `final`/`groundedness`, before any `visual`). `pending` is how many concepts are about to be
+     *  rendered (or null when not yet known). Use it to show a "generating visual(s)" state. */
+    onVisualsPending?: (p: VisualsPending) => void;
     /** A visual, emitted (out of anchor order) during the trailing visual stage when
      *  visual.mode="on". Place by `anchor`, not arrival order. */
     onVisual?: (v: Visual) => void;
@@ -561,6 +574,7 @@ export declare class AdminClient extends HttpBase {
             identity_prompt?: string;
             response_prompt?: string;
             generation_model?: string;
+            generation_model_mode?: string;
             max_response_tokens?: number;
             top_k_retrieved_chunks?: number;
             top_k_reranked_chunks?: number;

package/dist/index.js

@@ -364,6 +364,9 @@ function dispatchSse(frame, h) {
         case "final":
             h.onFinal?.(data);
             break;
+        case "visuals_pending":
+            h.onVisualsPending?.(data);
+            break;
         case "visual":
             h.onVisual?.(data);
             break;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@babav/knowledge-core-client",
-  "version": "0.22.1",
+  "version": "0.22.3",
   "description": "TypeScript client for the Babav Knowledge Core API (Deno + Node 18+, zero deps). Includes the babav.visual grammar TYPES at the ./visual subpath (types only; all visual rendering is server-side).",
   "type": "module",
   "main": "./dist/index.js",

package/src/index.ts

@@ -72,6 +72,7 @@ export interface RetrievalContent {
 
 export interface QueryOverrides {
   generation_model?: string;
+  generation_model_mode?: string;   // "reasoning" (adaptive thinking) | "standard" — how the TEXT answer is generated
   max_response_tokens?: number;     // max answer tokens
   top_k_retrieved_chunks?: number;  // candidate pool pulled from hybrid search (recall net)
   top_k_reranked_chunks?: number;   // sections kept after rerank -> the LLM
@@ -90,6 +91,7 @@ export interface QueryOverrides {
 /** Per-request visual overrides (see VisualRequest). */
 export interface VisualOverrides {
   concept_model?: string;
+  concept_model_mode?: string;  // "reasoning" | "standard" — how visual concepts are generated; OVERRULES generation_model_mode when combining
   concept_vet_model?: string;
   max_revisions?: number;
   claims_check_model?: string;
@@ -125,6 +127,13 @@ export interface Visual {
   image?: { url: string; expires_at: number; width: number; height: number };
 }
 
+/** Visual stage (streaming): generation has started. MAY be emitted twice — first with
+ *  `pending: null` (started, count unknown), then again with the ACCURATE post-vet count before
+ *  rendering. Take the LATEST value; `visuals_complete.count` is always authoritative. */
+export interface VisualsPending {
+  pending: number | null;
+}
+
 export interface VisualsMeta {
   count: number;
   dropped: number;
@@ -268,6 +277,7 @@ export interface Agent {
   identity_prompt: string | null;   // answer-system: who/purpose (null => server default)
   response_prompt: string | null;   // answer-system: output guidelines (null => server default)
   generation_model: string | null;
+  generation_model_mode: string | null;  // "reasoning" (adaptive thinking) | "standard"; per-agent
   max_response_tokens: number | null;
   top_k_retrieved_chunks: number | null;
   top_k_reranked_chunks: number | null;
@@ -455,6 +465,10 @@ export interface StreamHandlers {
    *  (native path). Also present in `final`. May be `{}` if scoring was unavailable. */
   onGroundedness?: (g: Groundedness | Record<string, never>) => void;
   onFinal?: (d: QueryResponse) => void;
+  /** First event of the trailing visual stage — visual generation has STARTED (fires after
+   *  `final`/`groundedness`, before any `visual`). `pending` is how many concepts are about to be
+   *  rendered (or null when not yet known). Use it to show a "generating visual(s)" state. */
+  onVisualsPending?: (p: VisualsPending) => void;
   /** A visual, emitted (out of anchor order) during the trailing visual stage when
    *  visual.mode="on". Place by `anchor`, not arrival order. */
   onVisual?: (v: Visual) => void;
@@ -681,7 +695,7 @@ export class AdminClient extends HttpBase {
 
   // Agents are query agents, global for now.
   agents = {
-    create: (b: { name: string; identity_prompt?: string; response_prompt?: string; generation_model?: string; max_response_tokens?: number; top_k_retrieved_chunks?: number; top_k_reranked_chunks?: number; sibling_window?: number; rerank_instruction?: string; max_subqueries?: number; max_reasoning_rounds?: number; decomposition_model?: string; reasoning_inspect_model?: string; groundedness_verifier_model?: string; groundedness_threshold?: number; grounding_enabled?: boolean; citations_enabled?: boolean }) =>
+    create: (b: { name: string; identity_prompt?: string; response_prompt?: string; generation_model?: string; generation_model_mode?: string; max_response_tokens?: number; top_k_retrieved_chunks?: number; top_k_reranked_chunks?: number; sibling_window?: number; rerank_instruction?: string; max_subqueries?: number; max_reasoning_rounds?: number; decomposition_model?: string; reasoning_inspect_model?: string; groundedness_verifier_model?: string; groundedness_threshold?: number; grounding_enabled?: boolean; citations_enabled?: boolean }) =>
       this.request<Agent>("POST", "/v1/agents", { json: b }),
     update: (id: UUID, b: Partial<Omit<Agent, "id">>) =>
       this.request<Agent>("PATCH", `/v1/agents/${id}`, { json: b }),
@@ -708,6 +722,7 @@ function dispatchSse(frame: string, h: StreamHandlers): void {
     case "citation": h.onCitation?.(data as Citation); break;
     case "groundedness": h.onGroundedness?.(data as never); break;
     case "final": h.onFinal?.(data as QueryResponse); break;
+    case "visuals_pending": h.onVisualsPending?.(data as VisualsPending); break;
     case "visual": h.onVisual?.(data as Visual); break;
     case "visuals_complete": h.onVisualsComplete?.(data as VisualsMeta); break;
     case "error": h.onError?.(data); break;