Skip to main content
The database · source-available

JouleDB.
The knowledge substrate for energy-optimized AI.

One energy-aware engine for transactional, analytical, and semantic workloads. Phasor-encoded hyperdimensional computing gives O(1) similarity, a deterministic-first substrate keeps work cheap, and every query carries a Merkle-anchored energy receipt.

3-in-1 OLTP · OLAP · semantic
O(1) similarity via HDC
BSL 1.1 source-available
jouledb · semantic search ⚡ 0.002 J
  • > "energy-aware schedulers"
  • exact-key probe L0 · index 0.000 J
  • HDC bind + bundle L1 · phasor 0.001 J
  • rank by resonance, O(1) L1 · phasor 0.001 J
12 results · Merkle-anchored receipt 0x9f3a…

A query, metered. Similarity in O(1).

The substrate

One engine, every workload

Transactional, analytical, and semantic, with energy as a first-class column.

Hybrid by design

Transactional, analytical, and semantic search in one engine. No bolt-on vector store, no copies shuttled between systems.

O(1) similarity with HDC

Phasor-encoded hyperdimensional computing turns similarity into a constant-time bind-and-bundle, not a brute-force scan over every vector.

Energy receipts, anchored

Every query carries a Merkle-anchored receipt of the joules it spent. Auditable energy, measured, not estimated.

The discipline

Your data layer, on a SWaP-2C budget

A database is pure SWaP-2C: how much hardware it pins, how much power it burns per query, what it costs to run, and how much heat it throws. JouleDB optimizes the whole budget.

Size

One engine, not a stack of specialized stores.

Weight

No copies shuttled between OLTP, OLAP, and vector systems.

Power

O(1) similarity and a deterministic-first path cut joules per query.

Cost

Fewer systems to license, run, and move data between.

Cooling

Less compute per query means less heat to pull out of the rack.

Build on the knowledge substrate

Source-available under BSL 1.1. jouledb.org →