AI Research Lab · Est. 2024

Where AI meets
quantum systems.

North9 builds AI systems at the infrastructure layer and runs the world's only independent quantum hardware verification above 40 qubits. Our research spans kernel-level ML, post-quantum cryptography, and exact quantum simulation.

200q
Max circuit size, exact
3.88×
IBM Eagle off from theory
0
Approximations, ever
MIT
Open-source license

Independent Findings We tested IBM Eagle and Google Willow

Classical simulators break down at ~40 qubits. Above that, there has been no independent verification of quantum hardware — until now. qRoute computes exact observables up to 200 qubits with error_bound=None. Here's what we found.

Hardware Verification Results · Surface Code · p_phys = 0.30% Exact theory computed by qRoute · 2024
IBM Eagle · d=3 · 17 physical qubits
Claimed logical error rate 0.2500%
Exact theory (qRoute) 0.0644%
3.88×
worse than theoretical optimum
Gap is unexplained by IBM's published noise model
Google Willow · d=7 · 97 physical qubits
Claimed logical error rate 0.0051%
Exact theory (qRoute) 0.0054%
<5%
from theoretical optimum
Google Willow independently confirmed accurate

Both verifications computed by qRoute — no truncation, no approximation, no classical simulator approximation artifacts. IBM Eagle's 3.88× discrepancy points to correlated errors, leakage, or measurement crosstalk not captured in their published noise model. Nobody else has had this view above d=3.


Product Quantum Hardware Verification

The standard validation method — comparing hardware output against a classical simulator — fails above ~40 qubits. North9 is the only organization that can independently verify quantum hardware at real operating scales. We compute the exact ideal result, then measure precisely how far your hardware is from it.

200q
Maximum circuit size computed exactly — no approximation, no truncation
0
Other independent verifiers at this scale exist anywhere in the world
100%
Reproducible, documented methodology — suitable for publication and due diligence
01
Submit your circuit
Send the circuit specification and the observables you're measuring. We support standard quantum circuit formats (QASM, Cirq, Qiskit). No proprietary lock-in.
02
qRoute computes the exact ideal
We compute the exact ideal result using proprietary methods. No Monte Carlo sampling, no matrix product state truncation. error_bound=None on every output.
03
You receive the exact gap
A full report: ideal result, your hardware's output, and the exact discrepancy broken down by observable and error type. Correlated errors, leakage, and crosstalk are all surfaced.
04
Independent, publication-ready
All methodology is documented and fully reproducible. Results are suitable for publication, investor due diligence, regulatory compliance, or competitive benchmarking.
Pricing Verification packages
Benchmark
$14,500
Single circuit, full discrepancy report. Right for academic groups and enterprise teams running a first independent validation.
  • 1 circuit, up to 100 qubits
  • Exact ideal result computation
  • Full observable-level gap analysis
  • Written methodology documentation
  • Results in 5 business days
Get Started →
Enterprise
Custom
Ongoing verification program for quantum hardware companies, cloud providers, and enterprises with continuous hardware access or regulatory requirements.
  • Unlimited circuits, full 200q range
  • Continuous verification cadence
  • Dedicated researcher point of contact
  • White-label or co-branded reporting
  • API access for automated benchmarking
  • NDA and co-authorship available
Contact Us →

All packages include full methodology documentation. Results are reproducible independently. Questions? hello@north9.org


Research Three active research directions
🔬
Exact Quantum Simulation

qRoute computes exact quantum observables up to 200 qubits — no truncation, no approximation. 18 published experiments spanning quantum foundations, error correction, drug discovery, and cryptography. The only tool of its kind above 40 qubits.

🧠
Kernel-Level AI

Neural Kernel embeds trained INT8 MLP models directly inside Linux kernel subsystems using eBPF — replacing handcrafted heuristics in the process scheduler, TCP congestion controller, and memory prefetcher with learned models that hot-swap at runtime.

🔐
Post-Quantum Cryptography

Production-grade cryptographic primitives engineered to survive quantum adversaries. ML-KEM-768 + X25519 hybrid KEM, Noise protocol handshakes, ChaCha20-Poly1305 AEAD. Deployed in Seam — a secure UDP transport library available today.

Open Source north9 — AI agent runtime

11 composable tools for building autonomous AI agents that run safely, remember everything, and cost less to operate.

🏗️
Stable
north9

Core runtime — Docker sandbox + persistent memory. Every command runs in isolation; structured state survives /compact and session restarts.

Docker MCP Memory
🔭
Stable
Lens

Observability — traces every tool call, token cost, and latency via PostToolUse hook. Query call counts and latency across sessions.

SQLite Hooks Tracing
🗂️
Stable
Index

Semantic memory — keyword search across all past agent runs. FTS5 BM25 ranking, tag filtering, persistent SQLite store.

FTS5 BM25 SQLite
⚒️
Stable
Forge

Eval framework — YAML test cases run against any model. Catch regressions before they reach production agents.

YAML Evals CI
🔐
Stable
Vault

Encrypted secrets — Fernet + PBKDF2 credential store. Inject API keys at runtime without exposing them in prompts or logs.

Fernet PBKDF2 Secrets
Stable
Grid

Parallelism — N tasks at 30s each still takes 30s total. Fan out work across threads with full result aggregation.

Threads Parallel Async
💰
Stable
Budget

Cost enforcement — hard token and dollar limits. Prevent runaway agent spend before it hits your billing dashboard.

Tokens Cost limits Guards
🚧
Stable
Gate

Policy enforcement — block dangerous tool calls before they execute via PreToolUse hook. Glob + regex rules, configurable YAML policy.

PreToolUse Policy YAML
🌐
Stable
Scout

Web memory — fetch any URL, extract clean text, search cached content by keyword. Agents browse without burning context on HTML noise.

HTTP FTS5 Cache
📊
Stable
Sift

Data queries — load CSV/JSON into SQLite, query with SQL. No need to read entire files into context; get exactly the rows you need.

SQLite CSV SQL
🔗
Stable
Chain

Workflows — YAML pipelines connecting all north9 tools. Steps reference each other with {{ step.output }} template syntax.

YAML Pipelines Orchestration
Install the full suite: pip install "git+https://github.com/North9-Labs/north9.git#egg=north9[mcp]" && python3 -m north9 --suite — requires Docker
Open Source Other tools

Ready to verify your hardware?

Whether you're building quantum hardware, running enterprise quantum programs, or conducting investor due diligence — this is the only independent verification that works at real scale.