Workspace
Explore files produced by Neat workflows without exposing arbitrary host paths. APIs support folders, bounded search, text preview, raw images, and virtual paths into MPK tar archives using archive::member syntax.
Get the active workspace root
Reports the logical root used by all workspace endpoints and whether the conventional `/workspace` mount is active. Returned paths establish context for the browser but subsequent API requests must still use relative workspace paths.
List a workspace folder or MPK archive folder
Lists the immediate children of a workspace directory or a directory inside an MPK tar archive. Use an empty path for the root and `archive_mpk.tar.gz::member/path` for archive navigation. Results are sorted and classified so clients can build a lazy-loading file browser without extracting archives.
Search indexed workspace paths
Performs a bounded, case-insensitive search across workspace filesystem names and relative paths. MPK archives are searchable as files, but members inside them are not indexed; browse an archive with the workspace tree endpoint instead. Results are ranked and capped at 200 entries. Supply `kind` to narrow broad queries to models, images, code, text, archives, executables, or binary files.
Get workspace path metadata
Returns classification and filesystem metadata for one file, folder, archive, or MPK member without reading its content. Use this when a client needs size, modification time, or preview strategy before requesting the file-preview or raw-image endpoint.
Get raw workspace image content
Returns original image bytes for an ordinary image file in the workspace filesystem. MPK archive-member paths are not supported by this endpoint. This route is intended for browser `<img>` sources and binary consumers; use the JSON file-preview endpoint for text and metadata. Non-image or inaccessible paths return an error rather than arbitrary file bytes.
Preview a workspace file or MPK archive member
Returns node metadata together with a bounded text preview when the selected content is text-like. The `previewAvailable` and `truncated` flags tell clients whether content was returned and whether only a prefix is shown. Binary files remain discoverable but are not serialized into the JSON response.