Pipelines and execution
Developer preview
Connect typed module ports while keeping authoring and execution status distinct.
Build workflows in Python
Create typed module definitions and pinned graphs programmatically, validate/export bundles and run finite Python workflows locally. Use ordinary caller-scoped catalog actions to save and publish versions.
What runs today?
| Path | Current scope |
|---|---|
| Python CLI | Explicitly trusted local modules with bounded contracts and new outputs. |
| Fixed inspection worker | Local video metadata inspection, durable attempts and accepted reports. |
| Organization machines | Queued runs of an immutable pipeline snapshot, pulled by enrolled machines with leases, logs and results. Implemented locally; hosted rollout pending. Trusted code in bounded processes, not a sandbox. |
Keep attempts and evidence separate
A run identifies its input snapshot, module/runtime version and attempts. Accepted outputs are published only after the applicable validation. Reanalysis creates new outputs.
The target managed design uses PostgreSQL for domain state and a trusted dispatcher for isolated CPU/GPU jobs. A provider status alone must never publish an unvalidated result.