What you build with it

Fifteen apps you could actually ship, each next to the stack you would reach for today — and exactly where Arkeion is stronger. Where the incumbent still wins, we say so. Numbers are from the repo's own benches.

Regulated & audited systems

2.51× faster durable writes than SQLite carrying the same audit guarantees

Patient records for a clinic (EHR)

Instead of Postgres + audit triggers + a history table

	Arkeion	Postgres + audit triggers + a history table
Proving no record was altered	SHA-256 hash chain + verify(), built into storage	A history table a privileged user can still edit — no cryptographic proof
Reading a record as it stood a year ago	SELECT … AS OF that version or timestamp	Replay the audit log by hand to reconstruct it
PII encryption at rest	AES-256-GCM per page, keys held by you	Bolt on pgcrypto or disk encryption, configured separately

Invoicing that must survive an audit

Instead of MySQL + an application-level audit log

	Arkeion	MySQL + an application-level audit log
The immutable trail	The storage IS the trail (append-only copy-on-write)	A log table the app writes — and can forget to write
Corrections	A new version; the old invoice stays addressable	UPDATE overwrites; history only if you logged it
Backup	Copy one file — the audit trail comes with it	mysqldump plus keeping the log table consistent

Chain-of-custody for legal evidence

Instead of Postgres + a WORM bucket + manual hashing

	Arkeion	Postgres + a WORM bucket + manual hashing
Detecting tampering	Every commit SHA-256-chained; verify() + anchors	Hash files yourself, store the hashes elsewhere, hope they match
Time-travel to any point	Native AS OF	Restore a backup to inspect the past
Bit-rot on cold storage	Reed-Solomon per page repairs it; scrub() walks the file	Bucket checksums detect corruption but do not repair it

AI, RAG & semantic search

14–21× smaller vector index than pgvector at 1M vectors

A support bot answering from your internal docs

Instead of Postgres + pgvector + Redis

	Arkeion	Postgres + pgvector + Redis
Vector + keyword in one query	Hybrid RRF (BM25 + vector) natively	Query pgvector and a text index, then merge in app code
Index footprint at 1M chunks	~39 MB (IVF-PQ)	551–820 MB (pgvector)
Keeping stores in sync	One file, one write path	App keeps DB + vector store + cache consistent

Semantic search over a product catalogue

Instead of Elasticsearch + a vector plugin

	Arkeion	Elasticsearch + a vector plugin
Systems to operate	One embedded engine	An Elasticsearch cluster to run, secure and scale
Filters + vectors together	SQL WHERE + ANN in one query	Combine the query DSL with kNN, tune both
At tens of millions of vectors	Single node; edge narrows (IVF is O(N) candidates)	A dedicated graph engine (HNSW) pulls ahead at scale

A notes app that finds by meaning

Instead of SQLite + a hosted vector DB

	Arkeion	SQLite + a hosted vector DB
Where vectors live	In the same file, next to the notes	A separate vector service alongside your SQLite
Offline / local-first	Fully embedded, no network call	The vector DB is usually a service round-trip
Computing the embeddings	External model — bring your own	External model too (Arkeion adds no ML deps)

In-app full-text search

0 extra search services to run

A help-centre / knowledge-base search box

Instead of Algolia (hosted search)

	Arkeion	Algolia (hosted search)
Where your content lives	In your database, on your infrastructure	Shipped to and stored in Algolia’s cloud
Cost model	Embedded — no per-search fee	Per-search / per-record pricing
Instant-search feel	Fast, self-hosted BM25 + snippet/highlight	Purpose-built: typically faster and more typo-tolerant

Search across a docs site or wiki

Instead of Elasticsearch

	Arkeion	Elasticsearch
Operations	Embedded, zero extra services	A JVM cluster to run and tune
Searching old versions	MATCH … AS OF searches the past	Reindex snapshots yourself
Raw query speed at scale	Single node	Scales out further for very large corpora

papaya — an email client, built on Arkeion

Built on Arkeion · in the works

Instead of SQLite + FTS5

	Arkeion	SQLite + FTS5
Search index encryption	Encrypted with the mail (AES per page)	The FTS5 index sits on disk in the clear
Searching old or deleted mail	MATCH … AS OF searches the past	Gone once the row is overwritten
Raw query & build speed	Index smaller on disk; competitive	Decades-mature C — still faster to query and build

Multi-tenant SaaS & the edge

~2× durable writes vs SQLite on real disk — one file per tenant

quota — API quotas, config & request logs at the edge

● Running on Arkeion in production

Instead of Postgres + Redis

	Arkeion	Postgres + Redis
Systems at each edge node	One embedded engine — it replaced both	A Postgres and a Redis to run and keep in sync
Config-change history	Versioned + AS OF, hash-chained	A separate audit table, if you built one
Durable log writes on real disk	~2× (append-only, one fsync per commit)	Two systems, two failure modes to reason about

svitrio — a multi-tenant control panel

Instead of Postgres schema-per-tenant

	Arkeion	Postgres schema-per-tenant
Tenant isolation	One file per project — physical	Shared cluster — logical isolation
Backup / offboard one tenant	cp or rm one file	pg_dump filtered by schema; deletes are delicate
Per-tenant encryption keys	Native — one key per file	Hard on a shared instance

A local-first desktop app

Instead of SQLite (plain)

	Arkeion	SQLite (plain)
Undo / history	Built-in versions + AS OF	Implement journaling yourself
Sync-friendly diffs	Branch + three-way merge	Diff rows by hand
Maturity & footprint	Young, pre-1.0	Decades-proven, smaller default footprint

Versioned data & migrations

3-way diff & merge on your data, not just your code

Feature flags with instant rollback

Instead of Postgres + a config table + audit

	Arkeion	Postgres + a config table + audit
Rolling back a bad change	AS OF / revert to a version instantly	Restore a backup or reverse the audit rows
Who changed what, provably	Hash-chained history	Trust an updated_by column
Try before you apply	Branch, test, merge	A staging copy and a manual promote

A headless CMS: edit on a branch, publish when ready

Instead of Postgres + a draft/publish workflow

	Arkeion	Postgres + a draft/publish workflow
A draft as a real branch	Data branch + merge	Draft columns or duplicated rows
Preview any past publish	AS OF	Keep versioned copies yourself
Diff two versions	diff_versions()	Write the diff query by hand

Versioned ML datasets you can diff and roll back

Instead of Parquet on S3 + DVC / lakeFS

	Arkeion	Parquet on S3 + DVC / lakeFS
Reading a version	SQL AS OF — query it directly, no checkout	Check out a commit, then read the files
Granularity of a diff	Row-level: changes() / diff_versions()	File-level, not per row
Petabyte-scale columnar analytics	Single-node, row-oriented	Built for huge columnar scans

See where the numbers come from

Every figure is benched against SQLite, pgvector, Qdrant and FTS5 — including where they win. Read the benchmarks →

Read the source