MonceApp v0.1.0

MonceApp: Context-Enhanced GenAI for Industrial Applications

Charles Dana, Monce SAS — April 2026 — v1.0

Abstract

MonceApp is a GenAI platform with 14 live models and 3 lazy constructors (Charles, Moncey, Json), serving 100 concurrent users on a $15/month EC2. The SDK (pip install monceai) requires zero API key. Arithmetic and glass decomposition bypass LLMs entirely via regex → boolean circuits → SAT solvers. The system integrates 10 computation services, uses Snake SAT for routing, and binary multiplier/divider circuits for zero-token exact arithmetic.

1. The (monceai-) Input Layer

Pure function: str prompt → str enhanced_prompt. Injects factory-specific glass knowledge, IGU rules, coating normalization, and operator context. Frameworks registered via POST /v1/framework.

2. The Model Layer — 14 Models, 3 Constructors

pip install git+https://github.com/Monce-AI/monceai-sdk.git

ConstructorTypeWhat
Charles("6x7")strMath, science, general — lazy parallel, resolves on read
Moncey("44.2 feuillete")strGlass sales agent — snake + classifiers + Haiku
Json("list 5 primes")dictStructured output — dict subclass, json.dumps(indent=2)
Json("..." + Moncey("..."))dictChains: Moncey resolves, Json structures
ModelEngineCost/msgLatency
charles4x parallel (haiku mem + csv + cnf + sudoku) → Sonnet~$0.018-15s
charles-scienceHaiku features → Snake router → 7 services → Sonnet~$0.0115-60s
charles-aumaHaiku encode → AUMA {0,1}^n → Haiku (= or ≈)~$0.0033-8s
concisecharles → Haiku TL;DR (420 tokens)~$0.0110-20s
cccharles ∥ concise → Haiku synthesis~$0.0212-25s
charles-jsonMemory (5s cap) → Sonnet strict JSON, VLM~$0.015-15s
charles-architectMemory → Sonnet ASCII diagrams~$0.015-15s
Sonnet 4.6Bedrock Converse + tools~$0.031-3s
Sonnet 4Bedrock Converse + tools~$0.032-4s
Haiku 4.5Bedrock Converse + tools~$0.0031-2s
Nova ProBedrock Converse (context only)~$0.0080.8s
Nova LiteBedrock Converse (context only)~$0.0010.7s
Nova MicroBedrock Converse (context only)~$0.00050.6s

3. charles-auma: 4 Compute Engines, 0-2 Haiku Calls

charles-auma has three execution paths, fastest wins:

Binary Multiplier Circuit (Zero LLM)

A×B encoded as carry-chain adder CNF: fix A,B bits as unit clauses, Kissat propagates to product bits. Division is the inverse: fix A,B, solve Q×B + R = A for Q and R.

ProblemAnswerVarsTimeTokens
6×7422797ms0
123×45656,08818755ms0
65535×655354,294,836,22576858ms0
42÷76 r 04582ms0
1000÷7142 r 69549ms0
998001÷999999 r 035348ms0

AUMA Boolean Search (Fourier Probing)

ProblemAnswerBitsAUMAEvals
6×7= 426bits [1,0,1,0,1,0]1,296
999²= 998,00120exact160,000
40-bit= exact40exact2,560,000

Polynomial Root Finding

PolynomialAUMA foundExactTime
x³ − 6x + 4 = 0x ≈ 0.7330.7320.6ms
x² − 7x + 12 = 0x ≈ 2.9830.4ms
z² + 1 = 0z ≈ −0.005 + 0.992i±i3.3ms
z&sup5; = 1z ≈ 1.0031 (principal)3.4ms

4. charles-science: Snake SAT Router

A Snake classifier (42 training samples, 15 layers, trained in 0.01s) routes prompts to the right service. Haiku extracts 16 boolean features from the prompt, Snake classifies instantly (0ms, 100% confidence on all tested cases), then Haiku extracts parameters for the selected service.

ServiceURLWhatLatency
AUMAauma.aws.monce.aiUniversal maximization, polynomial budget<100ms
Sudokusudoku.aws.monce.aiPolynomial BCP solver0.1-500ms
NPDollarsnpdollars.aws.monce.aiSwarm SAT solver5ms-60s
SnakeBatchsnakebatch.aws.monce.aiDistributed Snake training0.6-20s
Chesschess.aws.monce.aiAUMA chess engine vs Stockfish50ms-6s
Snakesnake.aws.monce.aiGlass article SAT matching0.15ms
KPIdata.aws.monce.aiLive extraction accuracy/volume2-5s

5. Moncey: Glass Industry Sales Agent

Deterministic-first, LLM-second. Like NP calc for arithmetic, snake/comprendre decomposes glass without LLM.

from monceai import Moncey, Json

Json("Extract order: " + Moncey("44.2 Silence/16 alu/4 rFloat JPP"))
# → {"articles": [{"name": "Feuillete 44.2", "ref": "#60442"}, ...]}
StepServiceWhatLatency
1snake.aws.monce.ai/comprendreGlass decomposition: verre, intercalaire, gas224ms
2moncesuite.aws.monce.ai/comprendre10 classifiers (only if snake quality < 75%)206ms
3HaikuFrench sales response with context~3s

6. The (-monceai) Output Layer

Snake SAT Memory: glass terms matched against local Snake classifiers (~50ms). NP-Complete Arithmetic: all formulas intercepted and computed exactly via AST-safe eval. charles-auma skips NP calc — AUMA is the calculator, finding answers from bits.

6. Connection to AUMA and the Dana Theorem

Snake SAT classifiers are built on the Dana Theorem (2024): any indicator function over a finite discrete domain can be encoded as a SAT instance in polynomial time. Snake is also the router — classifying prompt features into service selections at zero token cost.

charles-auma implements the AUMA thesis directly: any function f: {0,1}n → R can be maximized by Fourier probing + greedy search in O(na) evaluations. Integer arithmetic is a special case where the answer's binary representation IS the {0,1}n point that maximizes −(decoded − target)².

References

  1. Dana, C.M.J. (2023). "A O(2^n/2) Universal Maximization Algorithm." MSc Thesis, Ecole Polytechnique.
  2. Qadir, Dana et al. (2025). "Supervised ML identifies impaired mitochondrial quality control in beta cells." bioRxiv/Springer.