NEW RESEARCH: Your Sandbox Is Made of Glass
Read
Pricing
Every enterprise's AI risk surface is different. Your auditors, your insurers, your regulators, and your board are each measuring a different exposure — and a packaged plan cannot answer all four.
Trinitite's commercial structure is built around three measurable inputs: your inference volume, your policy complexity, and the audit and insurance posture your stakeholders demand. Book a conversation and we'll quote against those — not against a tier you have to grow into.
Measured Performance
0.00%
Governance variance under load
Batch-Invariant CUDA Kernels (Fixed-Size Tiling / Split-KV Lock). Identical inputs produce identical decisions regardless of GPU batch size or concurrent traffic.
100%
Malicious payloads blocked
All 500 attack payloads across five red-team scenarios — Context Poisoning, Streaming Payload, PII exfiltration, prompt injection, and unauthorized tool calls.
~404ms
Mean governance latency
Tested with Qwen3-0.6B deployed on batch-invariant SGLang. Standard deviation 0.0577s — versus 96.7s for Claude Sonnet under load.
0.29%
Of underlying inference cost
Governor operates at less than one third of one percent of inference spend, enabling 90%+ OpEx reduction via model downgrade arbitrage.
What Every Engagement Includes
There is no feature gating between "Free" and "Enterprise" on the Trinitite platform. Determinism, audit admissibility, and population coverage are not premium add-ons — they are what makes the platform legally defensible in the first place.
✓
Batch-invariant deterministic inference
0.00% drift regardless of server load, batch size, or GPU hardware. The same input always produces the same governance verdict.
✓
Glass Box Ledger with deterministic replay
Cryptographically signed State-Tuple Ledger that meets the Daubert standard for federal-court admissibility. Bit-for-bit reconstruction of any historical decision.
✓
400+ pre-built tool governors
Pre-trained governors for Stripe, AWS, GitHub, Salesforce, Slack, Zendesk, and every major SaaS surface your agents touch. Add custom governors for your domain.
✓
MCP Governance Server
A single proxy between every agent and every MCP server. Pass/correct/block verdicts on every tool call, multi-step attack-chain detection, credential redaction.
✓
Semantic Rectification
Geometric Policy Manifold maps unsafe outputs to the nearest compliant intent via Hilbert Space Projection. Business workflows continue; violations get fixed.
✓
Bitwise AGRC Framework mapping
11 domains, 78 controls — the Continuous Attestation Standard. Maps directly to EU AI Act, NIST AI RMF, ISO/IEC 42001, SOX 404, HIPAA, and SOC 2 Type II.
✓
Risk & Analytics — full module
Agentic Risk Score per agent, Monte Carlo loss simulations, VaR at 95%/99%, stress testing, loss triangles, RMIS integration, and the Broker Portal.
✓
Continuous Cryptographic Attestation
Real-time evidence pipelines for auditors, regulators, and insurers. No more questionnaire-based assurance. 100% population verification — never sampling.
How We Price
A published price list assumes every enterprise's exposure looks the same. The 4,000-iteration Trinitite red-team study proves it does not — safety drift, attack surface, and audit lift vary by an order of magnitude across customers. Pricing should vary with it.
01
Inference volume
Every AI tool call in your environment is intercepted, governed, and logged. Your commercial structure is sized to the actual transactional footprint of your agents — not the seats on your team.
02
Policy complexity
Baseline governors cover PII, credentials, prompt injection, and content safety. Custom-trained governors fine-tuned on your regulatory regime, proprietary schemas, and bespoke fairness constraints are scoped to the domains you actually operate in.
03
Audit & insurance posture
RMIS loss-run integration, Broker Portal access for underwriters, premium-tower modeling, IBNR reserve release analytics, and on-premise deployment options are priced against the audit and insurance lift they unlock for your specific stakeholders.
Pay only for what you sign
Start small. Add streaming when you trust the numbers. Turn on enforcement when your team is ready. Pricing scales with the volume you actually sign.
Layer 1 · Sampled Assurance
Pay per audited log.
Free first pre-audit. Signed in 72 hours.
Start
Free 1,000-log pre-audit. No card. No engineering work.
Pay as you grow
$0.04 per audited log over your annual bundle.
You walk away with
A signed gap report and a verify bundle anyone can re-run.
Layer 2 · Continuous Assurance
Pay per event you sign.
Ingest everything. Audit what counts.
Stream every event
$0.005 per event ingested through your connector.
Audit every event you sign for
$0.02 per event scored against your policy.
You walk away with
One signed pass rate per day. Drift webhook the day your model moves.
Layer 3 · Runtime Enforcement
Priced by governance calls.
Stop the next mistake — inline.
How we charge
Annual governance calls — sized to your AI traffic.
Latency SLA
900ms p99 entry tier down to 400ms p99 at scale.
You walk away with
Real-time block / correct / mask — with a signed receipt per action.
Framework library
15 frameworks pre-trained. SOC 2, HIPAA, EU AI Act, ISO 42001, NIST AI RMF, GDPR, CCPA, PCI-DSS, GLBA, FFIEC, SR 11-7, NYC LL 144, EEOC, Colorado AI Act, PCAOB AS 1215.
Custom framework
Audit against your internal policy. Built into your tenant's auditor LoRA once. Runs forever — no per-audit charge.
Bring your own GPU
Host the auditor on your infrastructure. Your data never leaves your network. Security review drops from 60 days to 14.
What Happens After You Book
Every Trinitite engagement starts with a measurable artifact, not a sales deck. You will leave the proof-of-concept with a liability-shielded figure expressed in dollars — built against your actual workloads, your actual policies, and your actual regulators.
Day 0
30-minute discovery call
We map your current agent fleet, model providers, MCP servers, regulatory regime, and the specific liability surface your board is asking about.
Days 1–7
Proof-of-concept against your real workloads
A scoped Trinitite environment intercepts a live slice of your agent traffic. You see exactly which tool calls would have been blocked, which were autocorrected, and what the Glass Box Ledger looks like for your data.
Days 7–30
Board-ready ROI model
A liability-shielded figure in dollars, expressed against your specific Intervention Density Ratio. The Monte Carlo VaR and stress-test scenarios are built against your actual workload — not a generic case study.
Day 30+
Commercial structure sized to actual usage
One contract. No per-seat math. No surprise overages. Pricing scales with your inference volume and the audit posture you actually need — and the Risk Decay Curve is measurable from week one.
The clock is already running
As of January 2026, ISO Endorsement CG 40 47 explicitly excludes ungoverned GenAI liability from standard Commercial General Liability policies. The EU AI Act's high-risk obligations are now in active enforcement. SOX 404 auditors are asking for population-level evidence of AI control effectiveness — not statistical samples.
Every day an ungoverned AI agent runs in production, your enterprise accumulates units of unpriced shadow liability. The Trinitite Governor converts that exposure into bounded, insurable, auditable evidence in 30 days. The longer you wait, the larger the figure you have to explain to your board.
Pricing FAQ
Trinitite is priced on usage, not seats. There are three metered layers: sampled assurance at $0.04 per audited log, continuous assurance at $0.005 per event ingested plus $0.02 per event scored, and runtime enforcement priced by annual governance calls sized to your AI traffic. Every engagement includes the full platform — determinism, deterministic replay, 400+ tool governors, the AGRC framework mapping, and the Risk & Analytics module — with no feature gating between a free pre-audit and enterprise.
A risk intelligence subscription is ongoing access to Trinitite’s Risk & Analytics module: a live Agentic Risk Score per agent, Monte Carlo loss simulations, Value-at-Risk at the 95th and 99th percentiles, stress testing and loss triangles, RMIS loss-run integration, and the Broker Portal for underwriters. It ships as the self-hosted Risk edition or as part of a SaaS subscription, so risk stays a continuously updated number rather than a quarterly report.
Editions decide which modules ship: Guardian covers runtime governance (MCP, NHI, sandbox); Audit covers AI assurance and evidence; Risk covers risk, insurance, and actuarial analytics; Enterprise bundles all modules self-hosted under a signed license; and SaaS turns everything on in the multi-tenant cloud. Most teams start with one metered layer — sampled assurance for a first signed gap report — then add streaming and inline enforcement as trust grows.
Thirty minutes. Your data. Your regulators. A real number you can take to the board.
30-minute discovery · Proof-of-concept against your real workloads · Board-ready ROI model
Trinitite
AI governance that catches mistakes, proves compliance, and shows the board what it saved—in dollars.
Trinitite is built by Fiscus Flows, Inc.
Products
Products
Solutions
© 2026 Fiscus Flows, Inc. · All rights reserved
Accessibility
The Guardian Standard™