VectorMethods

Showing 25 of 48 from the live probe.

• Find Matching Video Segments

  Find matching segments in videos attached to this message. This starts a fresh Playground workflow: upload the supplied videos, generate extraction schemas and optional filters together, and execute across exactly those videos. Transcription and image embeddings are off. Video segmentation uses content-aware processing. Prefer files forwarded by the host. Omit files only when the host cannot forward the videos; an inline picker will collect them. Local stdio clients can supply file_paths. Poll

• Upload Media

  Upload attached media through a durable import job. Prefer host-provided files. Omit files only when the host cannot forward attachments; an inline picker will collect them. Local stdio clients can supply file_path within allowed upload roots. Destination defaults to Playground; index_id selects an existing index and index_name reuses or creates a private index. Poll get_import_job with the returned job_id. To upload and immediately find matching segments, use find.

• Define Prompt

  Turn a natural-language instruction into a VideoVector prompt using Prompt Lab. The same generation call returns extraction schemas and an optional selection_filter only when the user wants to find, filter, or shortlist matching media. The generated instructions reflect supported segmentation intent, such as concise social clips or useful context around a target moment. The generated prompt is saved by default and the returned prompt_id can be passed directly to process_media. Set save=false on

• Process Media

  Process media with one prompt using a minimal target contract. Provide prompt_id or an inline prompt_instruction. With no target, the entire Playground is processed. Provide index_id/index_name for an index, or video_ids for selected media. Omitted execution settings inherit the engine configuration. Video and audio support fixed or content_aware segmentation, with content_aware as the default. The prompt may include an optional post-extraction selection filter. Returns a run immediately; poll

• Search Media

  Search processed media without first navigating VideoVector's internal abstractions. Use query for semantic vector search or filters for conditional search. With no scope, search the Playground. Scope by index, media IDs, or prompt-run IDs. Pass only cursor to retrieve the next frozen page without rerunning or rebilling the search.

• Search Videos

  Search for video segments using natural language. Returns relevant clips with timestamps, thumbnails, and extracted metadata. Use this tool when the user wants to: - Find specific content, moments, or topics in their video library - Search for scenes matching a description - Locate videos containing certain objects, actions, or concepts The search uses semantic similarity, so natural language queries work best.

• Search Videos by Image

  Find video segments that are visually similar to a reference image. Use this tool when the user has an image and wants to find similar content in their videos. The image must be provided as a base64-encoded string. Returns segments ordered by visual similarity, including the matched frame timestamp.

• Run Multimodal Search

  Search using both text and image queries combined. Uses Reciprocal Rank Fusion (RRF) to merge results from both modalities. Use this tool when the user wants to find content matching both: - A text description (what the content is about) - A visual reference (what it looks like) You can adjust weights to prioritize text or image similarity.

• Filter Videos

  Filter video segments by metadata field values using exact matches, ranges, or contains queries. Use this tool for precise filtering when: - You need exact matches (e.g., object_type = "vehicle") - You need numeric ranges (e.g., confidence > 0.8) - You need to check if an array contains a value Supports up to 4 filter conditions combined with AND logic.

• List Indexes

  List all video indexes the user has access to. Use this tool to discover available video collections before searching. Returns both user-created indexes and shared demo indexes. Each index contains videos that have been processed with specific prompts, enabling semantic search across their content.

• Get Index

  Get detailed information about a specific video index. Returns the index name, creation date, and whether it's a shared demo index.

• List Prompts

  List available analysis prompts that can be used to process videos. Prompts define what information to extract from videos (e.g., objects, actions, scenes). Each prompt has a JSON schema that defines the output structure. Use this to discover what analysis capabilities are available before running process_media.

• Get Video

  Get detailed information about a specific video. Returns video metadata including title, overall status, media type, duration_seconds, timestamps, and per-prompt-run segment processing snapshots (pending/processing/successful/failed).

• Get Video Segments

  Get all segments for a video with their extracted metadata. Segments are time-based chunks of the video with associated metadata from prompt analysis. Use this after search to get full details about a video's content.

• List Prompt Runs

  List prompt runs for the current user, a specific index, or a specific video. Use this to discover prior prompt executions before drilling into status, results, video-level summaries, billing settlement, or failed segments.

• Estimate Prompt Run

  Estimate the billing cost of a prompt run using the same configuration as execution. This does not start processing. Video/audio targets must already have server-measured duration metadata; otherwise the backend returns a duration validation error. Use it before process_media when the user wants pricing or balance confirmation.

• Get Prompt Run Status

  Check the status and progress of a prompt execution run. - Overall status (pending, processing, completed, completed_with_failures, failed, cancelled) - Progress counts (completed/partial/failed/cancelled media and segments) - Billing settlement fields (billing_status, billing_error, billing_estimated_mt, billing_actual_mt) - Any error messages Use this to monitor async processing jobs started with process_media.

• Cancel Prompt Run

  Request cancellation for a prompt run. Already-started videos may finish, but no new videos will start after the stop is observed. Returned run data includes billing_status and billing_error so callers can distinguish released reservations from pending settlement or billable partial cancellation. Use this when the user wants to stop an in-flight run because intent changed.

• Get Prompt Run Results

  Read bounded filtered and unfiltered extraction pages for a prompt run. Defaults to both pages across the entire run. Optionally narrow to one video_id. Each page has its own cursor; choose view=filtered or unfiltered to advance one page independently. result_level=segment returns segment metadata; video returns media-level synthesis. selection_summary distinguishes pending, ready (including zero matches), failed, and not_applicable. A null page is unrequested or unavailable. Poll get_prompt_ru

• Get Prompt Run Video Result

  Get the video/audio-level synthesis result for a specific media item inside a prompt run.

• Get Failed Prompt Run Segments

  Get the failed segment manifest for a prompt run grouped by video and operation.

• Retry Prompt Run Segment

  Retry a failed segment inside a completed prompt run without creating a new run. Returns retry billing fields. Terminal retry states settle billing_status to confirmed or released unless billing_error indicates an unresolved owner-aware billing operation.

• Get Segment Retry Status

  Get the status of an async prompt-run segment retry. Includes billing_status and billing_error for retry reservation settlement.

• Create Prompt

  Create a new custom extraction prompt with a JSON schema. Prompts define what information to extract from video segments. The JSON schema specifies the structure of extracted metadata. Use this tool when the user wants to: - Define custom metadata fields to extract from videos - Create a new analysis template for their specific use case - Set up structured data extraction for their video library The prompt_text is the instruction given to the AI model during extraction. The json_schema must b

• Get Prompt

  Get detailed information about a specific prompt including its full JSON schema. Returns the complete prompt definition with: - The prompt text (AI instruction) - Full JSON schema for extraction - Active status and creation date Use this to understand what a prompt extracts before using it with process_media.