Platform

DB

Live

AI-native distributed database layer

Open DBExplore ecosystem
189,065
Durable writes/sec
2.3 ms
BM25 p99 @ 5M docs
33
Workspace crates
2,870
Test functions

Overview

An open-source (Apache-2.0) Rust database unifying documents, BM25 search, vectors, and realtime in one engine. Benchmarks are published with committed evidence files — including the adversarial cases where recall degrades. Its Database 2050 roadmap tracks 100 capabilities, each with a pass/fail evidence gate.

CategoryPlatform
StatusLive
Domaindb.kynetra.dev

Built with

RustAxumTokioHNSWBM25Fjall LSM

Database 2050 roadmap

100 capabilities, each with an explicit evidence gate. A capability counts as implemented only when its gate passes on the main branch. The work below is built and passing internal gates on review branches — it is not merged or released yet.

26
Implemented
30
Partial
34
Planned
10
Research
In independent review

Consistency budgets and session tokens

Opaque, project-bound consistency tokens for read-your-writes, per-read freshness budgets, and replicas that forward tenant traffic to the primary through a bounded, streaming proxy.

In independent review

Transaction history checker

Searches recorded histories for G0, G1, G-single and G2 anomaly cycles with a bounded search — and reports “unverifiable” rather than a false pass when the bound is reached.

In independent review

Restore-proven backups

A backup only counts once a restore to a fresh target verifies its manifest and watermarks. Adds incremental schedules, point-in-time restore, cross-region object copies, automated restore drills, and a retention-locked vault.

In independent review

Crypto-agile encryption and key rotation

Envelopes record algorithm and key version, the write-ahead log derives a fresh key per record, and rotation drills enumerate every secret class — reporting any class that cannot rotate yet instead of skipping it.

In independent review

Adaptive query governor

Live resource budgets bound scans, fan-out, and result sizes for every request.

In independent review

Provenance for AI retrieval

Retrieval returns receipts naming source commits and ranking stages, vector indexes expose embedding freshness, and recall is fenced to a single index generation so a reindex can never mix results.

In independent review

Versioned formats and data contracts

Explicit, compatible format envelopes for exports, plus executable data contracts that enforce compatibility, ownership, and privacy rules at boundaries.

In independent review

Migration planning and two-way sync

A migration risk planner that inventories source shape, unsupported semantics, and rollback steps, and a bidirectional sync contract with explicit conflict semantics.

In independent review

Governance by default

Continuous sensitive-data classification, portable compliance evidence packs with credential redaction, attribute- and purpose-based authorization, policy as code, field-level lineage, and workload identity with mTLS.

In development

Data movement

Online schema evolution, a zero-downtime migration pipeline (snapshot, tail, verify, shadow-read, cutover), parallel bulk load and exit with checksums, and a federated query gateway.

Counts from the capability catalog on main (crates/server/resources/database-2050-capabilities.json) as of 26 September 2026. Nothing listed below is counted as implemented until it merges with its gate passing.

More in Platform

Prime

Self-learning regent intelligence OS

PrimeOS

100-layer AI-native operating system framework

Drive

Safety-bounded retrofit intelligence for existing cars

Auth

Unified SSO & identity management