Archive — old stuff
These pages describe earlier design phases of the project — most of them predate the actual implementation of the Distributed Composable Neural Runtime (DCNR) in the compute-all repository, and the implementation took different turns (typed neural VNodes instead of a generic operator set, orchestrator-managed networks instead of task dispatch, an S3-backed dataset API instead of a standalone Data Service).
They are kept here for historical context and because some of the ideas (validation strategies, trusted agents, data references) remain relevant for the upcoming permissionless-membership and trust work.
For the current state of the system, see:
- Architecture Overview
- Virtual Node (VNode) & Node Types
- Network Definition (YAML)
- Production Training Networks
- Project Status