Research Overview Tracks Agenda
Features Architecture Insights Research Overview Research Tracks Research Agenda Quick Start Roadmap Pricing Licensing Contact Documentation GitHub
Rust · Single Binary · Open Source

The Database That
Researches Itself

SkeinDB is a single-binary database combining cell-interned MVCC storage, MySQL corpus-backed compatibility, partial PostgreSQL support, 20 opt-in research features (18 hardened / 2 prototype), CDC changefeeds, and a full web admin — all in one Rust executable.

Get Started Read the Docs Support & pricing → View on GitHub →
147
RPC Methods
30+
Method Families
20
Research Tracks
27
Admin Panels
700+
Corpus Statements
No
100% Parity Claim

Core Capabilities

Everything ships in one binary. No external dependencies, no cluster of services.

Implemented
🧬

Cell-Interned MVCC

Every value is content-addressed. Identical cells across rows, versions, and tables share one physical representation — automatic deduplication without compression overhead. Segment-backed .rseg row storage is now the default serve mode.

Implemented
🔌

MySQL Wire Protocol

Connect with any MySQL client. SkeinDB speaks the MySQL binary protocol with 678 corpus-backed compatibility statements, window functions, user variables, prepared statements with projection-subquery metadata parity, broad COM_QUERY coverage, richer information_schema probes including plugins, partitions, constraints, parameters, tablespaces, privileges, and views, and WordPress admin parity including Users/Site Health query coverage, installer seed-query regressions, and a clean live admin sweep across the core screens.

Partial
🐘

PostgreSQL Wire Protocol

Connect with early PostgreSQL clients. SkeinDB now ships a partial PG v3 baseline on port 5432: trust/SCRAM auth, SSL rejection, simple and extended queries, transaction/savepoint state, virtual pg_catalog tables for tables/views, roles/users, indexes, tablespaces, sequences, statistics, and database health, plus common bootstrap probes like SHOW server_version and current_setting(...) backed by the shared engine.

Implemented
⚡

HTTP + QUIC Dual Transport

SkeinQL RPC runs over HTTP/1.1, HTTP/2, and QUIC from the same binary, with dedicated integration tests for transport parity and reconnect behavior.

Documented
📚

SkeinQL API Reference

The API surface now has a dedicated client reference: transports, envelopes, result formats, method families, and the current 147-method system.capabilities map are documented beside the full SkeinQL language spec.

Sample App
🧠

Vector RAG Retrieval

A new Vector RAG tutorial and samples/vector_rag_pipeline.py demonstrate deterministic embeddings, vector.insert, vector.search, and grounded prompt assembly without external credentials.

Signed
🔏

Signed macOS Artifacts

Tagged releases now build a macOS binary, seal it with codesign, verify the signature, and upload the archive, checksum, and signing metadata. The release packaging guide documents local Developer ID signing too.

Implemented
🖥️

Web Admin (SkeinAdmin)

27 interactive panels compiled into the binary, including a dedicated Help & Docs center with quick start, panel reference, R01-R20 research-track index, keyboard shortcuts, glossary, and direct links to the canonical documentation. Schema browsing, data manipulation, cluster control, dedicated telemetry/security centers, dialect-aware SQL workspaces, and dedicated panels for all 20 research features, including a click-first Easy Viewer with inline database creation plus live create-table preview/validation, expert control panels, a live Index Advisor page with ranked suggestions plus observed-before/expected-after scan reports, CDC lag and backpressure controls with pause/resume, average/p95/p99 query latency, plus per-fingerprint tail-latency histograms.

Prototype
🔬

20 Research Tracks (18 Hardened / 2 Prototype)

From learned ValueID indexes and differential privacy aggregate hardening to migration intent reports, NL-to-SkeinQL verification, and energy-aware compaction, 18 tracks are hardened with real algorithms and integration tests. R18 performance replay and R19 Wasm query operators remain prototype implemented while their remaining productionization work continues.

Prototype
🌐

Cluster Control-Plane

9 typed cluster methods for topology lifecycle: join tokens, node admission, shard placement, replica promotion, and RPC fanout replication with recursion suppression.

Architecture

Five layers from client interface to persistence, unified by typed SkeinQL method dispatch.

SkeinDB layered architecture

Operational Insights

Three runtime surfaces that show what SkeinDB is actually doing: manifest-backed shard movement, the embedded admin canvas, and compatibility depth across client entry points.

New in the cluster plane

Shard moves now expose transfer math before ownership flips

Rebalancing is no longer a blind metadata edit. SkeinDB now builds a shard-scoped object manifest from live rows, asks the destination which ValueIDs are missing, and only pulls the missing objects before changing primary placement.

Manifest Total objects and bytes are surfaced per move, not inferred after the fact.
Need-check The destination preflights ValueID coverage with objects.need.
Progress Batches, pulled objects, stored objects, and verification failures ride in the response.
Manifest-backed shard move progress surface
SkeinAdmin live dashboard surface
Embedded admin

Telemetry, security, CDC, cluster control, and research panels share one live control surface

SkeinAdmin is wired directly to the runtime. The dashboard, Easy Viewer, CDC pause/resume controls, index advisor, Privacy lab, Merge & CRDT evaluator, cluster tools, and Forensics proof bundles speak the same typed control plane as external clients.

27 admin panels 147 RPC methods 20 research tracks
MySQL and PostgreSQL compatibility surface
Compatibility depth

One engine answers MySQL wire clients, PostgreSQL bootstrap probes, HTTP RPC, and the admin stack

The same executable now carries a broad MySQL corpus, richer information_schema metadata for ORMs, PostgreSQL pg_catalog role/index/stat coverage, and the typed SkeinQL surface that powers both automation and the embedded UI.

678 MySQL corpus statements PG startup probes HTTP + QUIC transport

20 Research Tracks

Each track has RPC methods, a dedicated admin panel, and automated test coverage. Click any track for details.

R01Learned Index Structures R02Adaptive Row/Column Execution R03Delta-Chain Topology R04Differential Privacy R05Oblivious Execution R06Forensic Audit (WAL) R07Merge Functions (CRDT) R08Incremental View Maintenance R09QUIC-Native Protocol R10Vector Embeddings R11Autoparameterization R12NL → SkeinQL Translation R13Causal ETag Consistency R14Replay / Time-Travel R15Conflict-Free Schema Evolution R16Automatic Index Synthesis R17Migration Intent Inference R18Reproducible Performance Replay R19Wasm Query Operators R20Energy-Aware Compaction

Quick Start

From zero to running in four commands. Tagged releases also drive a repo-local Homebrew tap formula, signed macOS archives, and a signed apt publication path.

terminal 
# Clone the repository
git clone https://github.com/pinkysworld/SkeinDB.git
cd SkeinDB

# Build
cargo build --release

# Run all tests
cargo test

# Start the server
./target/release/skeindb serve --data ./data --http 8080

# Open the admin panel
http://127.0.0.1:8080/admin

# Send your first SkeinQL RPC request
curl -s -X POST http://127.0.0.1:8080/api/v1/rpc \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"system.ping","params":{}}'

Research Paper

Systems contribution covering architecture, cluster control, and 20 research tracks.

SkeinDB: A Single-Binary Database with Cell-Interned MVCC, a 20-Track Research Agenda, and a Web-Native Administration Stack

IJRCOM 2026 Systems Research Camera-Ready v5

Michél Nguyen · University of the People

SkeinDB addresses the divide between production database systems and research prototypes through a novel single-binary architecture that co-hosts a MySQL-compatible SQL layer, a typed RPC control surface (SkeinQL), and 20 experimental research features within one executable process. The paper presents four dimensions of contribution: cell-interned MVCC storage, persistent cluster control-plane, 20 research tracks with working implementations, and comprehensive web administration.


Author Profile (ORCID)
6
Professional Figures
18
IEEE References
6
Research Questions
7
Contributions

Built With

🦀
Rust
⚙️
Axum
🔒
Quinn (QUIC)
🐬
MySQL Protocol
🐘
PostgreSQL (partial)
📦
Single Binary