Skip to main content
Give the agent the workspace, project, editing brief, and intended destination. Ask it to identify candidates before reorganizing media.

Locate the project and folders

  1. Call workspaces_list to find workspaces available to the authenticated user.
  2. Call projects_list with the selected workspace_id.
  3. Call projects_get with the chosen project_ids. Its root_directory_id is the starting point for files at the top level.
  4. Use directories_list and assets_list with a parent_id to inspect the relevant folders. Follow pagination on tools that expose it.
Use IDs returned by these calls. A displayed project name is not a project ID.

Search using the available channels

Call assets_search_capabilities with the project_id before constructing a search specification. It describes the available visual, transcript, document, and filename channels and the metadata filters the project supports. Pass a supported spec to assets_search. Use asset_id when the search should be limited to one known asset. Discover valid option, Face, and user IDs instead of guessing filter values; entities_list helps find recognized people.
A matching transcript or indexed description is a lead for the edit. Check the actual clip before treating a match as a final select.

Read the surrounding context

Keep the asset ID and time range attached to each observation. Timecode conversion can return null with a reason when a source has no usable frame rate. Do not substitute invented timecodes.

Organize selects without moving the source

Create a named collection with collections_create, then add the chosen asset or directory IDs with collections_items_add. A collection groups references; it does not create local files or an editor timeline. Use fs_items_copy if the work needs separate cloud copies. It returns a background job ID. Use fs_items_move only when the intention is to change the source location; inspect reported name conflicts before supplying conflict resolutions. For metadata, discover fields with metadata_attributes_list, read current values with asset_metadata_get, and write supported values through asset_metadata_set. Reuse existing fields when they already describe the information.

Continue in an editor

The agent can now use the Aspect CLI to mount or download media and an editor integration to work with the timeline. Follow Agentic Video Editing to prepare mounted proxies and pin originals while cutting. For a visual planning board, read canvas_content_get before changing nodes or connections. Discover generation models with generations_models_list and inspect a run’s result before assuming a new asset is ready. See the tool reference.