Known limitations and non-goals (current)¶
This page is intentionally blunt. It exists so evaluators and operators can set expectations early.
Who this is for¶
- Evaluators deciding whether this suite fits their current stage
- Developers planning adoption and integration scope
- Operators who need to avoid “surprise missing feature” incidents
What you will get¶
- The current product boundaries: what is supported vs intentionally not included
- Concrete limitations that affect pilots and production rollouts
- Pointers to the relevant docs for workarounds and operational safety
Limitations¶
-
Tenant creation is DB-only today. There is no admin API to create tenants yet; bootstrap requires a SQL insert. See: Admin API + local bootstrap (recsys-service)
-
Pipelines manifest registry is filesystem-based.
recsys-pipelineswrites the “current manifest pointer” to its configuredregistry_diron the filesystem. Artifacts can be published to S3/MinIO, but publishing the manifest to object storage requires an explicit upload step (or serving from a file-based manifest template). -
Kafka ingestion is scaffolded. The
raw_source.type = kafkaconnector is present as a config option but is not implemented as a streaming consumer yet. -
Dev headers don’t carry roles. If you enable admin RBAC roles, admin endpoints require roles from JWT/API keys. For local dev with dev headers, either disable RBAC roles or use JWT.
Non-goals (by default)¶
- Running your infrastructure for you (managed hosting is not implied by this repo)
- “Magic” model training inside the serving stack (the suite is designed for deterministic, auditable behavior first)
Read next¶
- Pilot plan (what to do first): Pilot plan (2–6 weeks)
- Data modes (DB-only vs artifact/manifest): Data modes: DB-only vs artifact/manifest
- Local end-to-end tutorial (known-good baseline): local end-to-end (service → logging → eval)