@larkup/tool-video-intelligence 0.2.0

package/compose.yaml

@@ -0,0 +1,71 @@
+services:
+  video-intelligence:
+    build:
+      context: .
+      dockerfile: runtime/Dockerfile
+      target: cpu
+    image: ${LARKUP_VIDEO_IMAGE:-ghcr.io/larkup-ai/video-intelligence:0.1.0}
+    ports:
+      - '${LARKUP_VIDEO_PORT:-8787}:8787'
+    # Values are injected by the selected tool configuration at launch. This
+    # keeps a local runtime isolated from developer machine secrets.
+    environment:
+      LARKUP_VIDEO_DATA_DIR: /data
+      LARKUP_VIDEO_MODEL_DIR: /models
+      LARKUP_VIDEO_RUNTIME_KIND: local-docker
+      LARKUP_VIDEO_REQUIRE_AUTH: ${LARKUP_VIDEO_REQUIRE_AUTH:-false}
+      LARKUP_VIDEO_SHARED_API_KEY: ${LARKUP_VIDEO_SHARED_API_KEY:-}
+      LARKUP_VIDEO_ADMIN_TOKEN: ${LARKUP_VIDEO_ADMIN_TOKEN:-}
+      LARKUP_VIDEO_WORKERS: ${LARKUP_VIDEO_WORKERS:-1}
+      LARKUP_VIDEO_DEVICE: ${LARKUP_VIDEO_DEVICE:-auto}
+      # Local compute remains on the user's machine. The optional provider
+      # calls below use only the user's own selected API key.
+      LARKUP_VIDEO_SEMANTIC_VISION: ${LARKUP_VIDEO_SEMANTIC_VISION:-false}
+      LARKUP_VIDEO_VISION_PROVIDER: ${LARKUP_VIDEO_VISION_PROVIDER:-vercel_ai_gateway}
+      LARKUP_VIDEO_VISION_API_KEY: ${LARKUP_VIDEO_VISION_API_KEY:-}
+      AI_GATEWAY_API_KEY: ${AI_GATEWAY_API_KEY:-}
+      LARKUP_VIDEO_SEMANTIC_VISION_MODEL: ${LARKUP_VIDEO_SEMANTIC_VISION_MODEL:-alibaba/qwen3.8-27b}
+      LARKUP_VIDEO_REASONING_VISION_MODEL: ${LARKUP_VIDEO_REASONING_VISION_MODEL:-alibaba/qwen3-vl-235b-a22b-instruct}
+      LARKUP_VIDEO_AGENT_ENABLED: ${LARKUP_VIDEO_AGENT_ENABLED:-false}
+      LARKUP_VIDEO_AGENT_PROVIDER: ${LARKUP_VIDEO_AGENT_PROVIDER:-vercel_ai_gateway}
+      LARKUP_VIDEO_AGENT_API_KEY: ${LARKUP_VIDEO_AGENT_API_KEY:-}
+      LARKUP_VIDEO_AGENT_MODEL: ${LARKUP_VIDEO_AGENT_MODEL:-openai/gpt-5-mini}
+      LARKUP_VIDEO_PLANNER_MODEL: ${LARKUP_VIDEO_PLANNER_MODEL:-openai/gpt-5-mini}
+      LARKUP_VIDEO_TRANSCRIPTION_PROVIDER: ${LARKUP_VIDEO_TRANSCRIPTION_PROVIDER:-}
+      LARKUP_VIDEO_TRANSCRIPTION_FALLBACK: ${LARKUP_VIDEO_TRANSCRIPTION_FALLBACK:-}
+      DEEPGRAM_API_KEY: ${DEEPGRAM_API_KEY:-}
+      LARKUP_VIDEO_DEEPGRAM_MODEL: ${LARKUP_VIDEO_DEEPGRAM_MODEL:-}
+      LARKUP_VIDEO_DEEPGRAM_AUTO_MODEL: ${LARKUP_VIDEO_DEEPGRAM_AUTO_MODEL:-}
+      OPENAI_API_KEY: ${OPENAI_API_KEY:-}
+      LARKUP_VIDEO_OPENAI_TRANSCRIPTION_MODEL: ${LARKUP_VIDEO_OPENAI_TRANSCRIPTION_MODEL:-}
+      GROQ_API_KEY: ${GROQ_API_KEY:-}
+      LARKUP_VIDEO_GROQ_TRANSCRIPTION_MODEL: ${LARKUP_VIDEO_GROQ_TRANSCRIPTION_MODEL:-}
+      ELEVENLABS_API_KEY: ${ELEVENLABS_API_KEY:-}
+      LARKUP_VIDEO_ELEVENLABS_TRANSCRIPTION_MODEL: ${LARKUP_VIDEO_ELEVENLABS_TRANSCRIPTION_MODEL:-}
+      LARKUP_VIDEO_EMBEDDING_PROVIDER: ${LARKUP_VIDEO_EMBEDDING_PROVIDER:-disabled}
+      DASHSCOPE_API_KEY: ${DASHSCOPE_API_KEY:-}
+      DASHSCOPE_WORKSPACE_ID: ${DASHSCOPE_WORKSPACE_ID:-}
+      DASHSCOPE_REGION: ${DASHSCOPE_REGION:-}
+      RUNPOD_API_KEY: ${RUNPOD_API_KEY:-}
+      LARKUP_VIDEO_RUNPOD_EMBEDDING_ENDPOINT_ID: ${LARKUP_VIDEO_RUNPOD_EMBEDDING_ENDPOINT_ID:-}
+      LARKUP_VIDEO_HF_EMBEDDING_URL: ${LARKUP_VIDEO_HF_EMBEDDING_URL:-}
+      HF_TOKEN: ${HF_TOKEN:-}
+    volumes:
+      - video-intelligence-data:/data
+      - video-intelligence-models:/models
+    healthcheck:
+      test:
+        [
+          'CMD',
+          'python3',
+          '-c',
+          "import urllib.request; urllib.request.urlopen('http://127.0.0.1:8787/v1/health', timeout=4)",
+        ]
+      interval: 15s
+      timeout: 5s
+      retries: 10
+    restart: unless-stopped
+
+volumes:
+  video-intelligence-data:
+  video-intelligence-models:

package/dist/agent.d.ts

@@ -0,0 +1,131 @@
+import type { AgentToolExecutionContext } from '@larkup/marketplace/extension';
+import type { VideoIntelligenceClient } from './client.js';
+type InspectionPurpose = 'verify-visual' | 'high-res-ocr' | 'compare' | 'count' | 'track' | 'code';
+type AnalysisMode = 'fast' | 'balanced' | 'thorough';
+interface InspectVideoKnowledgeInput {
+    mediaAssetId: string;
+    startSecs: number;
+    endSecs: number;
+    purpose: InspectionPurpose;
+    queryId: string;
+    question?: string;
+    maxFrames?: number;
+    analysisMode?: AnalysisMode;
+    knownEntities?: string[];
+    continuousSequence?: boolean;
+    includeSpeech?: boolean;
+    maxWaitMs?: number;
+}
+interface QueryVideoEvidenceInput {
+    mediaAssetId: string;
+    query: string;
+    limit?: number;
+    exhaustive?: boolean;
+    cursor?: number;
+}
+export interface VideoIntelligenceAgentClient extends VideoIntelligenceClient {
+    queryVideoEvidence(input: QueryVideoEvidenceInput, context: AgentToolExecutionContext): Promise<unknown>;
+    inspectVideoKnowledge(input: InspectVideoKnowledgeInput, context: AgentToolExecutionContext): Promise<unknown>;
+}
+/** Chat actions stay inside the installed tool; the host only sees workflow roles. */
+export declare const AGENT_TOOLS: ({
+    name: string;
+    description: string;
+    parameters: {
+        type: string;
+        additionalProperties: boolean;
+        required: string[];
+        properties: {
+            mediaAssetId: {
+                type: string;
+                minLength: number;
+            };
+            query: {
+                type: string;
+                minLength: number;
+                maxLength: number;
+            };
+            limit: {
+                type: string;
+                minimum: number;
+                maximum: number;
+            };
+            exhaustive: {
+                type: string;
+            };
+            cursor: {
+                type: string;
+                minimum: number;
+            };
+            startSecs?: undefined;
+            endSecs?: undefined;
+            purpose?: undefined;
+            queryId?: undefined;
+            question?: undefined;
+            maxFrames?: undefined;
+            continuousSequence?: undefined;
+            includeSpeech?: undefined;
+        };
+    };
+    method: string;
+    workflow: "evidence-query";
+    evidenceInput: "media-asset";
+    systemPromptFragment: string;
+} | {
+    name: string;
+    description: string;
+    parameters: {
+        type: string;
+        additionalProperties: boolean;
+        required: string[];
+        properties: {
+            mediaAssetId: {
+                type: string;
+                minLength?: undefined;
+            };
+            startSecs: {
+                type: string;
+                minimum: number;
+            };
+            endSecs: {
+                type: string;
+                minimum: number;
+            };
+            purpose: {
+                type: string;
+                enum: string[];
+            };
+            queryId: {
+                type: string;
+                minLength: number;
+                maxLength: number;
+            };
+            question: {
+                type: string;
+                minLength: number;
+                maxLength: number;
+            };
+            maxFrames: {
+                type: string;
+                minimum: number;
+                maximum: number;
+            };
+            continuousSequence: {
+                type: string;
+            };
+            includeSpeech: {
+                type: string;
+            };
+            query?: undefined;
+            limit?: undefined;
+            exhaustive?: undefined;
+            cursor?: undefined;
+        };
+    };
+    method: string;
+    workflow: "evidence-refinement";
+    systemPromptFragment: string;
+    evidenceInput?: undefined;
+})[];
+export declare function attachVideoIntelligenceAgentClient(client: VideoIntelligenceClient, fetcher?: typeof globalThis.fetch): VideoIntelligenceAgentClient;
+export {};