01 QELM Core
The working public model framework.
QELM can create, train, save, load, and run compact language
models with quantum attention and feed-forward paths. The
public release includes dataset preparation, a trainer,
model serialization, token maps, a lightweight chat UI,
and local or IBM-compatible backend routes.
Available now
- PyPI package and public source
- Scalar and sub-bit encoding
- Parameter-shift, SPSA, and Adam training
- Trainer GUI and saved-model chat UI
Active work
- Full sequential multi-block execution
- Independent amplitude and phase learning
- Faster GPU and larger-model paths
- QSA 0.2 estimator and execution integration
Open QELM source →
02 Qubit State Algebra
A compiled exact runtime that plans around quantum structure.
QSA 0.2.0 is QELM's native quantum execution layer. Its C++20
runtime keeps independent and locally entangled work in the
smallest certified exact representation available, then adds
reusable operation, tensor, estimator, and gradient plans for
repeated workloads.
Current release
- QSA 0.2.0 with C++, C, and Python interfaces
- Reusable exact tensor and Pauli estimator plans
- Parameterized sweeps and exact adjoint gradients
- Linux, Windows, macOS, sanitizer, and reference gates
Exactness contract
- Specialized routes require certified structure
- Resource caps fail closed without hidden truncation
- QRegister remains the general exact fallback
- Published multipliers apply only to named workloads
Open QSA 0.2.0 →
03 Brain integration
A governed intelligence layer, not a public chat mockup.
Brain is the protected local system built around QELM's model
path. It separates operator control, retrieval, inference,
semantic action, background learning, and evaluation so that a
generated answer does not automatically become memory, authority,
or a promoted model update.
- 01Task contractScope, route, and authority
- 02Governed contextRecall, relationships, and provenance
- 03QELM executionModel path plus bounded QSA work
- 04Evaluation gateScore, retain, promote, or roll back
Current system work
- Persistent local service separated from the operator UI
- Semantic recall, structured relationships, replay, and consolidation
- Typed actions with validation, provenance, confidence, and authority
- Held-out checks, checkpoints, recovery, and rollback
Evidence status
- Current Brain/QELM integration is protected and active
- An internal checkpoint is hash-verified
- The public Brain repository is an outdated historical prototype
- Independent current-model validation is not yet published
Protected boundary: model weights, private corpora, memory contents, topology, thresholds, credentials, and security-sensitive protocols are not published.
Open the Brain research record →
04 Qubit network
A public local foundation; a distributed runtime still in development.
The public Qubit repository contains Qiskit/Aer-backed and
CPU-only logical qubit foundations. The forward architecture
is a temporary channel network where devices accept bounded
work, return results, and leave without holding Brain’s full
model or long-term memory.
Public today
- HybridQubit simulator foundation
- CPU-only Cubit implementation
- Logical subspaces and noise experiments
- Public architecture documentation
Not released
- Desktop node installer
- Mobile node application
- Public distributed lease service
- Production node security review
Open Qubit foundation →
05 Neuron
An early public learning-record experiment.
Neuron currently represents a small public starting point
for learning records and related telemetry. It is not a
production service, a replacement for Brain, or evidence of
a completed distributed neural fabric.
Current boundary
- Public early-stage repository
- Inspectable learning-record direction
- Separate from Brain’s protected authority
- Integration remains active research
What it may support
- Traceable learning events
- Run and update telemetry
- Comparative development records
- Auditable model-change history
Open Neuron →
