Product maturity and supported deployment scope

Current position

Spedo is a single-node, in-memory data engine in preview. It supports a

documented subset of RESP and is intended for evaluation, development, and

carefully qualified single-node workloads. It is not a drop-in replacement for

every Redis deployment, and it must not be presented as a highly available

database.

The snapshot/WAL reliability and crash recovery architecture is detailed in the official operator guide: persistence-and-recovery.md.

Before adopting a feature, validate the exact commands, data semantics, and

failure behaviour required by the application. The source of truth for the

release version is Cargo.toml; deployment defaults are defined by the active

Compose profile and src/main.rs.

Area	Status	Supported statement	Do not claim or rely on
Core key/value over RESP	Preview, single node	In-memory RESP operations covered by the documented/tested command subset.	Full Redis compatibility, multi-node durability, or automatic failover.
Python SDK (`spedo-client`)	Preview	`from spedo_client import SpedoClient` is the canonical public import.	A server ACK for local-cache or write-behind enqueue paths.
KV/JSON snapshots and WAL	Preview, single node	`SPDO2` checksummed checkpoints with previous-generation fallback; `SPEDO_WAL_FSYNC=always` synchronizes ACKed KV/JSON mutations to the local volume.	HA, quorum replication, remote backup, durability of Lists/Streams/vectors/etc., or a general database guarantee.
Vectors & Hybrid Search	Preview	SIMD AVX-512 vector similarity (L2, Cosine, Dot) and BM25 hybrid ranking on single node.	Distributed sharding or multi-tenant vector cluster.
Reactive State & CDC	Preview	In-memory `SpedoLiveVar` and stream-based change-data-capture.	Multi-datacenter global ordering without external coordinator.
Server-Side Rhai Scripting	Preview	Sandboxed in-memory Rhai script evaluation with AST caching and timeout guarantees.	Unrestricted host OS or filesystem access from scripts.
Support bundle	Preview tooling	`spedo-support-bundle` collects redacted read-only diagnostics for a reported incident.	A complete forensic export, an automatic root-cause analysis, or a guarantee that every proprietary secret format is detected.

redundancy, backup, and failover policy in the surrounding infrastructure.

SPEDO_USERS or SPEDO_REQUIREPASS before allowing remote connections.

Treat metrics, the admin UI, SAVE, SAVE WAIT, PERSIST.STATUS, and other control-plane operations as

privileged. Do not publish them through an Internet-facing load balancer.

persistence commitment.

Replication may be called HA only after initial synchronization, incremental

mutation transfer, resynchronization, promotion/fencing, and network-partition

tests are implemented and published. Tiered storage may be called durable only

after atomic writes, restart recovery, delete/tombstone handling, disk-full,

and corruption tests are published. Until then, the statuses above remain the

public product contract.