Category 5 data infrastructure

The world's first and only Category 5 database built from storage to mesh.

The algorithms are public. The architecture is not. 8DB makes encryption, transparent proofs, integrity, provenance and governance one database-native boundary—without multiplying cryptographic cost across every row.

*To our knowledge, based on public product documentation reviewed through July 18, 2026. Defined scope: at rest and the 8DB device mesh.

+0.054%full-profile integrity storage
0.24 msboundary signature verification
44×smaller than per-record signed PostgreSQL
8ONE SUBSTRATE
GRAPHVECTORMESHTABLEMEDIATIME
CATEGORY 5 BOUNDARY
ML-KEM-1024ML-DSA-87AES-256-GCM-SIV

Full-profile economics

Quantum-safe is available. Making it nearly free is not.

The measured Category 5 profile combines confidentiality, integrity, signing and tamper evidence. 8DB applies those controls at a native dataset boundary instead of repeating a 4,627-byte signature beside every record.

Full-profile storage100K × 115 B
44×smaller than per-record signed PostgreSQL
8DB
13.1 MB
PostgreSQL
579.4 MB
MongoDB
495.2 MB

Executed composed workload. Competitor figures use independent ML-DSA-87 signatures per datum—the path required when individual non-repudiation cannot be batched.

Integrity verification100K records
0.24 msone signed boundary
8DB
0.24 ms
PostgreSQL
160.3 s
MongoDB
159.8 s

8DB verifies one release-executed boundary signature. The per-record competitor path must verify 100,000 independent signatures. Expert batching is shown separately below.

+0.054%integrity storage

Full-profile dataset-boundary signing over the already encrypted 8DB state.

0 BPQC bytes per record

Classical and PQC key establishment use the identical 131-byte record path.

O(log N)datum inclusion proof

Verify a datum against the signed boundary without storing an independent signature beside it.

The buyer progression

Four protection states. One native path reaches the fourth.

PostgreSQL and MongoDB provide native classical encryption. Reaching a post-quantum, signed and tamper-evident state requires an application-layer integrity system. 8DB keeps that state inside the database boundary.

01

Plaintext

No protection. Useful only as the measurement baseline.

8DBAvailable for harness isolationPostgreSQLNativeMongoDBNative
02

Classical encryption

AES-protected data with classical key establishment.

8DBNativePostgreSQLNative / extensionMongoDBNative CSFLE path
03

PQC confidentiality

Quantum-safe establishment of the key protecting bulk data.

8DBNativePostgreSQLApp-layer onlyMongoDBApp-layer only

The expert alternative

Batching recovers bytes. It does not recreate the boundary.

PostgreSQL + Merkle batch24.5 MB0.19 s verify
MongoDB + Merkle batch19.4 MB0.73 s verify

Measured at 100,000 records. A competent app-layer Merkle design avoids per-record signature bloat, but it must independently maintain the authenticated tree, update and re-sign its root, produce inclusion proofs and remain atomically consistent with database mutations, provenance and governed keys. The 8DB advantage here is integrated capability—not a claim that batching is impossible.

Defense edge

Fit the protected evidence on the device.

At 100K telemetry records: 13.1 MB in 8DB versus 579.4 MB with individually signed PostgreSQL rows.

Regulated finance

Preserve individual attribution without signature bloat.

When per-transaction non-repudiation disallows batching, conventional stores repeat a 4,627-byte signature for every transaction.

PQC migration

Move now without changing the record economics.

Classical and PQC key establishment use the same per-record cipher path; the full-profile integrity layer adds 0.054% storage.

A protected data path

Protection begins with the first write.

Sealing, query and mesh integrity share one architecture. Each layer carries a precise, testable boundary.

Privacy by construction

Privacy that survives the whole data lifecycle.

8DB reduces the places where plaintext, reusable keys, stable identity, excess attributes and recoverable stale state have to exist. One scope model follows data through storage, access, proof, search, mesh, recovery and erasure.

Protect bytesDerive accessProve eligibilitySearch sealedSuppress tracesRecover or erase by scope
01

Born protected

PQC-established encryption protects storage and mesh transit. The default store exposes no default plaintext-at-rest path.

02

Unlock by policy

Authentication, scope and current policy derive ephemeral key material below the API. The wrong context cannot open the state.

03

Prove less

Transparent post-quantum ZK profiles can prove membership, eligibility or one-time use without revealing the record or hidden credential.

04

Search sealed

The protected index opens only inside the authorized boundary. Query leakage is named and measured instead of hidden by an umbrella claim.

05

Locked means absent

Search results, notifications, thumbnails, previews, embeddings and cached derivatives inherit the same privacy scope.

06

Recover by scope

Choose device quorum, guardians, enterprise recovery, a recovery kit—or no recovery—without one universal database backdoor.

One coherent boundaryEncryption hides the bytes. 8DB also minimizes keys, identity, attributes, traces and stale state.

Zero knowledge applies to defined proof-bearing operations. Policy-derived access, designated-verifier attestations, protected search, freshness and erasure solve different privacy problems—and compose inside the same substrate.

One source of truth

Every data modality. One protected substrate.

Not a finite multi-model catalog. 8DB supports every modality as a projection over the same governed atoms—including structures defined after the data is stored. Add a new way to understand the data without adding ETL, another database, a replication path or a security policy.

Example projectionsRepresentative, not exhaustive

Familiar modalities

8DB native structures

G

Familiar modalities · Graph

Relationships stay first class.

Entities, edges, provenance and policy live in the same protected substrate. Applications can move from a record to its context without exporting data into a separate graph service.

Best fitMission context, fraud networks, supply chains, biological pathways

Cross-modal proof

One question. Four modalities. No federated join.

Which vessels are linked to a target owner, entered the North Sea in the time window, triggered a sensor anomaly and meet the required confidence threshold?

No internal ETL between engines. No duplicated policy boundary. The repeated question can be maintained as a write-time intersection and resolved as one protected lookup.
2-hop graph+spatial+time series+confidence
20.3 µsfull intersection pipeline10-run Criterion mean · Apple M1

Performance, end to end

Fast is a system property.

A high-security database has to do more than win a microbenchmark. Buyers feel response time, sustained work, concurrency, query complexity, scale, storage traffic, energy and recovery. 8DB measures the whole operating profile—with protection active.

01 · Latency500 / 600 nsPQC write / read

Release, embedded path. Classical and PQC writes both measured 500 ns; reads measured 700 and 600 ns.

02 · Tail latency32.8 msp99 video ingest

Versus 159.2 ms for SQLite in the same Apple M1 media workload.

03 · Throughput27.6MYCSB Workload B ops/sec

Single-node c6a.8xlarge result; 8.9× Redis at matched high-pipeline load.

04 · Concurrency5.03×throughput at 8 readers

Eight concurrent readers completed 8× the work with 5.03× single-reader throughput.

05 · Query efficiency20.3 µscross-modal intersection

Graph + spatial + time series + confidence resolved without a federated join.

06 · Scale345.6Mmetrics committed in 8.19 s

TSBS wire + sort + page build + learned-index pipeline on a single c6a.8xlarge.

07 · Storage + I/O0 Badditional PQC bytes per record

The KEM envelope amortizes to 0.0123% at 100,000 records; measured engine WAF is 3.04.

08 · Energy3.8×more photos per charge

1.15M vs 303K in the measured Apple M1 media-ingest comparison with SQLite.

No consolidation tax

One engine. Specialist-class speed.

Supporting every modality only matters if buyers do not surrender the performance of databases built for one. Across established single-modality harnesses, 8DB meets or exceeds specialist systems in several workloads—and remains competitive where the specialist leads.

Key-value27.6Mops/sec

YCSB Workload B

7–9× Redis across measured workloads on the same c6a.8xlarge
Graph296.6Kqueries/sec

LDBC SNB IS-1

148× Neo4j 5.26 on the same c6a.8xlarge
Time series42.2Mmetrics/sec

TSBS full pipeline

9.6× the referenced QuestDB 2024 published record
Analytics1.92×faster

ClickBench 10-query subset

6 of 10 wins vs ClickHouse on the same c6a.4xlarge
Spatial2.51 sQ12 runtime

SpatialBench · 41M trips

205.1M returned rows · 7.34× faster than the R-tree baseline
Vector4,707queries/sec

VectorDBBench SIFT-1M

Recall@10 0.957 · #4 in the measured comparison

Representative measured results. Each comparison names its workload and scope; hardware, protocol, persistence path and approximation settings materially affect database benchmarks.

Embedded autonomy

Every node can keep knowing.

8DB lives inside the application or device. Local storage, query and decision context continue through disconnection. Sealed state synchronizes when peers return.

  • 01
    Software-onlyNo required HSM, accelerator or trusted execution environment; enclaves remain compatible
  • 02
    Locally authoritativeAir-gap ingest, storage and query remain available
  • 03
    Epistemic contextProvenance, confidence and policy can travel with local state
  • 04
    Recovery tested64 of 64 abrupt-stop trials recovered with zero corruption
01EDGE
02LOCAL
8SEALED
03PEER
04AIR GAP

NO CENTRAL DEPENDENCY

Buyer alternatives

Upgrade a layer—or replace the stack of layers.

PQ transport is becoming an incumbent feature. The difficult buyer problem is protecting storage, query, keys, edge operation and every data path as one authorized system.

Buyer decision8DBOracle / IBM / CockroachMongoDB QE / Always EncryptedSurreal / Ditto / CouchbaseAssembled stack
Post-quantum boundary PQ or hybrid TLS; scope and parameter set vary No comparable PQ database boundary documented No comparable end-to-end PQ boundary documented Achievable component by component
Authorization package Mature procurement; PQ transport covers one layer Point capability within a larger system boundary Point capability within a larger system boundary Every component and seam must be assessed
Search protected data Add protected-query layer Core strength, with query and leakage limits Add protected-query layer Add a specialist
Private verification Add proof system and state commitments Protected query is not a general proof layer Add privacy and proof layers Achievable with added provers, commitments and policy wiring
Lifecycle privacy Coordinate database, IAM, search, cache, backup and policy services Field and query protection; lifecycle controls remain external Add policy, proof and erasure closure Achievable, with the most seams to keep synchronized
Embedded offline mesh Server / cluster centric Server centric Strongest in Ditto / Couchbase Add edge database + sync
Multimodal + knowledge Strongest in Oracle; product dependent Narrower data model Strongest in Surreal; not the full intersection Multiple engines + ETL
Full-profile integrity economics Add a row-signing or external Merkle layer Add an app-layer signature / Merkle system Add PQ signing, proofs and governed keys Batching can recover bytes; every mutation seam remains yours
Security and operating surface Mature core plus added seams Drivers, key vault, KMS and query constraints Add PQ, query and knowledge layers Most identities, paths, copies and recovery seams
Adoption implicationLowest change if PQ transport is enough; accept coverage gapsProtect selected fields; accept query, schema and platform constraintsSolve data or edge need; add the remaining security layersPreserve chosen components; operate and authorize the whole composition
01

Adopt 8DB

Replace overlapping data and security layers with one protected substrate, fewer data copies and fewer integration boundaries.

02

Upgrade the incumbent

Keep operational maturity and migration compatibility when PQ transport is the actual requirement.

03

Assemble the stack

Reach broad coverage with known parts, while carrying the highest integration, attack-surface and authorization burden.

Best fit: the more a workload needs protection, query, edge autonomy and multimodality at once, the stronger the 8DB consolidation advantage becomes.

Built for consequential data

One architecture. Different mission pressure.

Defense

Operate through the disconnect.

Protect long-lived mission data at the tactical edge. Keep local systems running through denied or intermittent connectivity, then synchronize sealed state across the device mesh.

  • 01Autonomous platforms
  • 02Sensor and C2 data
  • 03Logistics and maintenance
Relevant proof

Category 5 algorithms, air-gap operation, anti-replay and crash recovery evidence

Evidence before adjectives

Built to survive diligence.

Every public number needs a boundary, an environment and a reproduction path. The limits are part of the proof.

01

Exact product boundary

Storage and device-mesh protection are packaged as one evaluated configuration, with public-edge profiles composed separately.

02

Defined query privacy

Protected search is measured, and each deployment selects an explicit query, key-custody and access-pattern profile.

03

Hardware independence

Core protection runs in software. Enclaves and HSMs remain optional defense or authorization choices.

04

Transparent verification

Named proof profiles expose their predicate, verifier and leakage boundary instead of using zero knowledge as a blanket claim.

Answers security teams ask for

Trust begins with precise answers.

How does 8DB support regulated authorization?+

8DB packages the cryptographic suite, key lifecycle, operating boundary, configuration and evidence needed to evaluate an exact deployment. Government authorization remains system- and mission-specific rather than attaching universally to a database product name.

Does 8DB require secure enclaves or an HSM?+

No specialized hardware is required for the core protection path. Rollback freshness can compare the local tamper-evident causal anchor with a peer-witnessed or published tip, providing the external monotonic reference in software. 8DB remains enclave-compatible for defense in depth, and customer authorization or key-custody policy may still require an HSM or validated module.

Does 0 B mean encryption has no storage overhead?+

No. Plaintext was 115 B and both encrypted records were 131 B. The measured claim is that PQC added zero further bytes per record versus classical encryption because both use the same AES-256-GCM-SIV data path. PQC adds one 1,616 B key-establishment envelope, equal to 0.0123 percent at 100,000 records in this harness.

Does 44 times describe every traditional database?+

No. It describes the executed 100,000-record, 115-byte fixture against PostgreSQL with an independent 4,627-byte ML-DSA-87 signature per record. An expert Merkle batch reduced PostgreSQL to 24.5 MB, versus 13.1 MB for 8DB, but required a separate application-layer mutation, proof and provenance boundary. Both alternatives are shown because the correct comparison depends on whether individual non-repudiation permits batching.

Is every 8DB privacy control zero knowledge?+

No—and conflating them would weaken the security model. Zero knowledge applies to named proof-bearing operations with a defined statement, witness and verifier. Encryption protects bytes; policy-derived keys enforce scoped access; designated-verifier attestations bind policy; sealed search reduces plaintext exposure; freshness resists rollback; and erasure closes the lifecycle. Each mechanism states what it hides, what it proves and what it can still reveal.

What does “locked” mean beyond denying a read?+

The same scope must close secondary surfaces: search and autocomplete, notifications, thumbnails and previews, embeddings, caches, analytics, replicas and other derived artifacts. That prevents a product from hiding the record while still revealing its existence or content through another subsystem.

Does recovery create a universal backdoor?+

It does not have to. 8DB recovery can be selected per scope: device quorum, guardians, a recovery kit, enterprise officer quorum or no recovery. The security review still covers share rotation, former guardians, accumulated compromise, coercion and cross-scope escalation.

How is this different from SurrealDB?+

SurrealDB is a strong modern reference for native multimodality, developer experience and embedded deployment. 8DB is positioned around the additional combination of Category 5 storage and device-mesh protection, protected query, decentralized edge operation and broader topological or hypergraph structures.

Does any database cover the full government PQC requirement by itself?+

There is no single universal government database certificate. Publicly documented incumbents cover pieces—primarily PQ or hybrid transport—while 8DB delivers the storage-to-mesh capability intersection in one product and one evaluation boundary. The authorizing body evaluates the exact deployed system.

Does a polyglot data stack create attack vectors?+

It can increase attack surface, but it is not insecure by definition. Each additional engine, connector and ETL path adds identities, credentials, network calls, serializers, logs, backups, patches, data copies and policy mappings that must be secured and monitored. A well-engineered zero-trust composition can manage these risks; it simply has more boundaries to prove and operate.

When is an incumbent the safer choice?+

When the requirement is limited to PQ transport, established certification and procurement maturity may outweigh architectural breadth. 8DB is most compelling when the full combination of high-assurance protection, edge autonomy, encrypted query and multimodality is required.

8Braid Journal

Follow the possibility. Inspect the proof.

Ideas, research, benchmarks and engineering notes about the protected information plane.

Explore the Journal
03research

Model-Independent Data

AI Is Outgrowing the Way We Store Information

Models are advancing faster than the information layer beneath them. 8DB is designed to let one protected fact participate in relational, graph, vector, temporal, spatial and media computation without becoming a collection of disconnected truths.

A better technical conversation

Evaluate the boundary, not the brochure.

Bring the workload, threat model and authorization target. Turn the claims into an acceptance plan your security and data teams can test.