Runtime modes
ABR mode, guest mode, persistent account mode and optional non-ABR learning systems.
Source map
- python/abria_server/api.py
- python/abria_server/abr_runtime.py
- python/abria_server/foundation_training.py
- python/abria_server/neural_v2.py
ABR mode
When ABR mode is active, foundation training endpoints are intentionally disabled for the primary reasoning path. ABR develops per-user cognitive fabric from experiences instead of requiring pretrained/foundation weights.
Guest sessions
Guests receive a deterministic guest-* id derived from the conversation session id. They are isolated from other sessions but their fabric is not saved across service restarts like authenticated accounts.
Authenticated ABR
A bearer session maps to an AETHER-backed account UUID. Persistent experience ingestion and personal fabric saving require this authenticated identity.
Optional neural and foundation modules
neural_v2.py, foundation_training.py, online_learning.py and trainer code still exist for experiments and non-ABR pathways. They should be documented as optional layers rather than described as the definition of ABR.