Architecture Overview¶

K-Pipe is a uv workspace of self-contained packages under packages/. Each package owns its full source tree under its own src/ (e.g. packages/core/src/kpipe/, packages/nuke/src/kpipe_nuke/) — there is no shared top-level lib//scripts// extensions//res/ tree. The core library (kpipe) lives in packages/core/src/kpipe/; optional DCC packages (e.g. kpipe_blender, kpipe_nuke) each bundle their own launcher and DCC extension together.

Key modules¶

Module	Location	Role
`config.py`	`packages/core/src/kpipe/`	Loads TOML configs; resolves per-user overrides (`<key path>.<username>.<key>`)
`entity.py`	`packages/core/src/kpipe/`	`Entity`, `EntityData` (metadata TOML), `EntityFileHelper` (naming)
`production.py`	`packages/core/src/kpipe/`	`ProductionManager` singleton; file-locked TOML cache (30-min TTL)
`attributes.py`	`packages/core/src/kpipe/`	Config queries — FPS, handles; project formats `get_format()` / `get_formats()` → `Format` (name, resolution, overscan)
`media/`	`packages/core/src/kpipe/`	DCC-agnostic media vocabulary: format/codec registries, settings, presets, OCIO color roles, frame ranges, file-type detection — see Media Settings & Presets
`publish.py`	`packages/core/src/kpipe/`	`publish_file/video/sequence/texture_set()` — copies outputs into the publish tree and records metadata, including the `media` settings (legacy `integrate_*()` functions are deprecated)
`pipeline.py`	`packages/core/src/kpipe/`	Five-phase publish workflow and action registry
`session.py`	`packages/core/src/kpipe/`	DCC workfile lifecycle with DCC-specific callbacks
`path.py`	`packages/core/src/kpipe/`	Template-based path resolution

DCC integration pattern¶

Each DCC follows a three-layer pattern:

Launcher script        →  resolves DCC install path, sets env vars, spawns DCC
  └─ Extension         →  startup scripts loaded by the DCC, initialises K-Pipe context
       └─ Session      →  manages active entity/task; DCC registers open/save/close callbacks

See DCC Integration for details.

Production backend¶

ProductionManager abstracts the data source via ProductionBackend. The only implemented backend is Kitsu (packages/kitsu/src/kpipe_kitsu/backend.py) using the gazu SDK. Results are cached locally in a TOML file with filelock for multi-process safety.

Set data_backend = "kitsu" (or "local") in the project config.

Shots in the cache carry their Kitsu casting: a casting list of {asset_id, asset_type, asset, count, label, animated} entries (animated is true for Kitsu's animate label and false for fixed). It is read with ProductionManager.get_shot_casting(sequence, shot) or Entity.get_casting(); asset_type/asset match the asset cache keys.