Constraint Intelligence Platform

Hard constraint decisions before the deadline.

Model + time budget → best verified assignment. Residual violations and provenance included.

Developer access via GitHub Sponsors — purchase API credits at 1.5× your sponsorship amount.

3.7 msSUTRA median latency
90.1%SUTRA complete-SAT
5,140SUTRA workload runs
113MNitroSAT V3 clauses
10MQ-State constraints

Distinct campaigns — open the evidence ledger.

Why teams switch

Replace expensive, opaque constraint bottlenecks.

Organizations lose time and money when constraint-heavy decisions take hours, fail without explanation, or cannot ship behind an API. Navokoj turns that bottleneck into a product surface.

Existing workflow
With Navokoj
8-hour solve cycle or overnight batch
+ Timeout-bounded solve with anytime feasible results
No useful explanation when infeasible
+ DEFEKT diagnostics and residual-violation reporting
Expensive seat or enterprise solver licenses
+ Monthly workload plans with transparent compute overage
Difficult integration into product backends
+ REST API with clear result semantics and docs
Binary success / failure only
+ Best-known assignment by deadline, with status flags
Rebuild and re-parse models from scratch
+ Incremental and persistent constraint workflow roadmap
RAM-bound large-instance workflows
+ Streaming / artifact-backed stores via NitroSAT V3 lineage
Claims scattered across prose and slides
+ Public evidence ledger, manifests, and verifier roadmap

Read the full switch narrative in Why customers switch to Navokoj.

Product contract

Model + deadline → verified decision.

Navokoj is an anytime constraint runtime. You do not wait for a black box to finish overnight. You get an inspectable result your application can act on.

01

Encode

Send Boolean CNF, weighted WCNF, schedule models, or finite-domain constraints that capture hard rules and soft preferences.

02

Bound

Attach a time budget and optional engine or hardware preference. The runtime optimizes under that deadline — not an open-ended batch.

03

Decide

Receive the best verified assignment available by the deadline, with status flags, residual violations, provenance, and billing metadata.

Result semantics buyers care about

  • solved=true — every clause is satisfied
  • feasible=true — all hard constraints are satisfied (soft may remain open)
  • Partial / timeout — best-known assignment returned with residuals
  • Provenance — engine, hardware, runtime, and billing in the response
Read the full API contract →
{
  "solved": false,
  "feasible": true,
  "assignment": { "nurse_3_shift_2": true, "...": "..." },
  "residual_violations": [],
  "runtime_ms": 148,
  "engine": "nitro"
}

Where it fits

Workloads technical buyers bring first.

Start with one expensive constraint decision. Measure baseline time, failure mode, and integration cost — then let the runtime earn a place in the stack.

Workforce & hospital scheduling

Shifts, skills, rest rules, and coverage must all hold — and the roster must ship before the next cycle.

You send
Schedule or CNF/WCNF model + deadline
You get
Hard-feasible roster with residual soft violations ranked
Nurse roster case study →

Logistics & allocation

Capacity, routing, and resource assignment under hard operational constraints.

You send
Boolean or finite-domain model of rules and objectives
You get
Best assignment available by budget, inspectable by ops
API docs →

Verification & policy

Configuration, protocol, or circuit constraints where “maybe” is not good enough for hard rules.

You send
CNF / hybrid models with verification metadata requests
You get
SAT/feasible result plus provenance for audit trails
Shipped API verification →

AI-agent planning & control

Agents need constraint-checked plans, not unconstrained tool loops.

You send
Planning or control-plane constraints as API workloads
You get
Deadline-bounded decisions agents can gate on
Constraint intelligence for agents →

Put hard decisions behind an API.

Send a model and a deadline. Get an assignment your application can inspect and act on.

Constraint Execution API

Integrate constraint execution into your product. Submit a model and a deadline; Navokoj returns a decision your application can inspect, verify, and act on.

  • Method: POST
  • Endpoint: /v1/solve
  • Input: Boolean, weighted, or finite-domain model
  • Output: Decision + verification metadata

Built for

  • Schedule: workforce and resource planning
  • Verify: configuration and policy checks
  • Decide: bounded, inspectable assignments
Use Case 1: Employee Scheduling (Python)
python
import requests

# 8 employees, 3 shifts, 7 days = 168 variables
payload = {
    "num_vars": 168,
    "clauses": [
        # Employee 1 must work at least one shift
        [1, 2, 3],
        # Cannot work morning AND afternoon (Mutually Exclusive)
        [-1, -2], [-2, -3], [-1, -3] 
        # ... more constraints
    ],
    "engine": "nitro"
}

response = requests.post(
    "https://api.navokoj.shunyabar.foo/v1/solve",
    headers={"Authorization": "Bearer YOUR_KEY"}, 
    json=payload
)

Evidence you can inspect.

Selected results are linked to their artifacts, configurations, and verification records.

SUTRA • 90.1% complete-SAT

SUTRA constraint workloads

Production SUTRA engine • interactive Boolean solving

Scope 5,140 workload runs
Result 90.1% complete
Headline 3.7 ms
Runtime Median latency
Evidence ledger →
Verified SAT

Hospital scheduling

Client-verified hard-feasible roster

Scope 504 variables
Result 39,842 clauses
Headline 100% hard-feasible
Runtime ~0.15s
Case study →
Streaming scale

Enterprise timetabling

Bounded-memory streaming execution

Scope 147,600 variables
Result 113M clauses
Headline 100% satisfied
Runtime 4m48s
Evidence ledger →
Finite-domain

Finite-domain planning

Q-State execution at industrial scale

Scope 100K nodes
Result 10M constraints
Headline Completed
Runtime ~8.66s
Evidence ledger →
Independently checked

Gate-level circuit verification

Exact refinement with independent checker

Scope 24,193 variables
Result 80,325 clauses
Headline Exact SAT
Runtime 13.76s
Verification notes →

Campaigns and configurations differ. Prefer the Hugging Face evidence ledger over isolated headlines.

Two Ways to Use Navokoj

Open Source Core + Production API

NitroSAT provides an open research and engineering release. Navokoj provides the managed runtime, routing, verification, and commercial execution surface.

Open-source SAT/MaxSAT research and engineering paths, including the V1, V2, and V3 solver lineage.

  • Run SAT and MaxSAT workloads locally
  • Work with large Boolean models and benchmark suites
  • Stream large formulas with bounded memory
  • Inspect assignments, results, and reproducibility artifacts
Best for
Research, prototyping, and local large-scale solving
76.28% perfect SAT rate • V3 streaming up to 113M clauses
PRODUCTION

SUTRA Production API

navokoj.shunyabar.foo

SUTRA is the production constraint engine behind the hosted runtime, with anytime results, diagnostics, verification metadata, and usage billing.

  • Boolean, weighted, and finite-domain workloads
  • Deadline-bounded best-known assignments
  • CPU, GPU, repair, and streaming execution paths
  • Verification-ready results and provenance
Production proven
92.57% Pro complete-SAT on SAT 2024 Industrial Track
CPU runs under 10k variables; GPU execution supports higher-variable workloads

What the production API has solved

116M clauses
Streaming timetabling
3.7 ms median
5,140 SUTRA workload runs
10M constraints
Finite-domain execution
100% verified
Hospital scheduling
13.76 seconds
Exact circuit SAT

Engine lineage

From solver experiments to a production runtime.

Each stage added a different operating capability. Today, the API exposes the production SUTRA path through the stable engine: "nitro" contract.

Nano

Lightweight first-pass solving for small, fast requests.

Mini

Balanced in-memory solving for general API workloads.

Pro

Higher-quality production solving for mission-critical workloads.

NitroSAT

Open-source SAT/MaxSAT research and engineering release.

V2

High-throughput in-memory optimization for formulas that fit RAM.

V3

Streaming, incremental, bounded-memory execution for large formulas.

SUTRA

The production constraint engine behind Navokoj’s stable API runtime.

See it in action

Product overview

Watch how Navokoj returns deadline-bounded, inspectable constraint decisions for production workloads.

Deterministic, time-bounded constraint solving

Continuous dynamics and physics-inspired search for over-constrained systems where binary success/failure is not enough — and the deadline is real.

Diagnostic intelligence

DEFEKT: pre-flight for constraints

Estimate difficulty before you spend a full solve budget. DEFEKT ranks likely conflict structure so you can route, debug, or avoid wasteful runs.

Cost control

Avoid burning GPU minutes on workloads that look hopeless under budget.

Constraint debugging

Rank likely conflict groups and residual violations for human review.

Smart routing

Choose engine and hardware suited to the model and deadline.

/v1/diagnose
Solvability score 84%
Status Likely solvable
Recommendation

Route to the engine and hardware best suited to this model and budget.

Start with a plan. Scale with compute.

Monthly workload entitlements, explicit rate limits, and usage-priced acceleration.

Hobbyist

Evaluate the runtime on small CPU workloads.

$0
forever
  • 300 SAT solves / month
  • 25 Q-State solves / month
  • 10 schedules / month
  • 10K variables and 100K clauses
  • 1-request account ceiling; CPU access
Get API access

Mini Lab

Build prototypes and low-volume production workflows.

$19
per month
$190 annually - two months included
  • 2,000 SAT solves / month
  • 250 Q-State solves / month
  • 100 schedules / month
  • 100K variables and 1M clauses
  • 3-request account ceiling; CPU access
Get API access

Lotus Fleet

High-volume operation across every compute class.

$499
per month
$4,990 annually - two months included
  • 50,000 SAT solves / month
  • 5,000 Q-State solves / month
  • 2,000 schedules / month
  • 2M variables and 12M clauses
  • 25-request account ceiling; CPU + L4 + H100
Get API access

Compute settlement

Plan access first. Hardware usage second.

CPU entitlement is included in each plan. Overage and GPU execution draw from API credits.

Included CPU Within monthly entitlement $0
CPU overage Up to 1M billable clauses $0.01 + $0.02/min
Large CPU Above 1M billable clauses $0.05 + $0.02/min
L4 GPU Usage billed $0.25 + $0.10/min
H100 GPU Usage billed $1.50 + $1.00/min

Dedicated capacity

More model size. More throughput. Your hardware.

Dedicated deployments are optimized for maximum model size and throughput. Limits are based on the hardware provisioned for your account rather than shared-cloud ceilings.

Dedicated deployments from $500/month. L4: from $500/month RTX 5090: from approximately $925/month Indicative 24/7 pricing: current GPU infrastructure reference cost plus $200/month. Final pricing depends on region, availability, and hardware. Talk to us about capacity

Private runtime

A premium deployment, not another shared tier.

  • Dedicated infrastructure with no multi-tenancy
  • Private API endpoint and private workload boundary
  • Usage on the provisioned machine without shared-plan solve quotas
  • No shared-cloud rate limits or concurrency throttling
  • Custom solver configuration and deployment support
  • Air-gapped operation available

Anytime runtime

A deadline returns a decision, not an empty wait.

Partial results include residual violations and provenance. Shared capacity has a 30-minute timeout; dedicated longer-running capacity is available by arrangement.

Developer access via GitHub Sponsors

Purchase Navokoj API compute credits through GitHub Sponsors. Sponsor ShunyaBar Labs to receive credits at 1.5x your sponsorship amount.

  1. Purchase credits via GitHub Sponsors.
  2. Email contact@shunyabar.foo with the transaction ID and your Navokoj account email.
  3. We provision or credit the account associated with that email.

Billing: compute time is measured per second. The API response and ledger are authoritative.

Canonical API contract

Plans, workloads, and hardware each set a limit.

The effective request limit is the lowest applicable value. This prevents a high-volume CPU entitlement from overrunning a reserved GPU queue.

Account plans

PlanMonthlySAT / moQ-State / moSchedules / moVars / clausesConcurrentHourly30 sec
Hobbyist$0300251010K / 100K11205
Mini Lab$192,000250100100K / 1M31,00015
Launch Pad$19910,0001,0005001M / 8M1010,00040
Lotus Fleet$49950,0005,0002,0002M / 12M2550,00080

Per-offering requests every 30 seconds

OfferingHobbyistMini LabLaunch PadLotus Fleet
Nano5154080
Mini261632
SUTRA (`engine: nitro`)3103060
Pro--410
Ensemble---2
Q-State24816
Scheduling1248
Diagnostics5102040
Batch requests1248

Hardware admission and queue ceilings

Constraint metricStandard CPUL4 GPUH100 GPU
30-second request ceiling6063
Concurrent request ceiling832
Boolean variables1,000,000100,0002,000,000
Boolean clauses8,000,000300,00012,000,000
XOR variables / constraints1,000 / 50010,000 / 5,000100,000 / 50,000
Q-State nodes / states1,000 / 1615,000 / 20100,000 / 50
Scheduling resources1002001,000

Account concurrency spans hardware pools; each pool also enforces its own ceiling. A batch consumes one batch-request unit and one solver, account, hardware, and hourly unit per submitted model. GPU access is plan-gated and usage-billed. GET /v1/pricing is the machine-readable source of truth for policy version, quotas, bursts, and prices.

Questions technical buyers ask

Result semantics, pilots, pricing, and verification—without the hype.

Developer access

Buy API credits. Ship a constrained decision.

Developer access via GitHub Sponsors: purchase Navokoj API compute credits through a developer-friendly payment rail. Sponsor ShunyaBar Labs to receive credits at 1.5× your sponsorship amount, then email the transaction ID and registration email for key provisioning.

Claim inbox: contact@shunyabar.foo · Careers: shunyabarlabs.github.io/careers