SALT Reliability · AIR · August 2026
The reliability platform for the AI era.
One product: AIR. Proven hardware FMEA, FRACAS, and CAPA disciplines, applied to AI systems that already sit in critical operations — with the evidence an auditor can read.
Problem
AI is in critical operations. The reliability record is not.
No engineering framework
There is still no standard way to apply FMEA, FRACAS, and CAPA to models, agents, and evals. Teams keep the story in tickets, slides, and memory.
Documents about AI, not from it
Generic GRC tools make you write policies. They do not mint the audit chain from the work of operating the system.
The mandate already exists
DoD TEVV has been in force since 2022. NIST AI RMF is the civilian counterpart. EU AI Act Article 50 transparency obligations apply from August 2026. The tooling to operationalize any of it does not.
Solution
AIR is the closed loop for AI reliability.
AI-FMEA
Typed failure modes: hallucination, drift, adversarial manipulation, tool-contract failure, waste. S/O/D scoring survives. Detection maps to eval coverage and monitoring.
Incident management
Red-team findings, drift events, guardrail breaches, eval regressions, and field failures share one lifecycle: report, triage, RCA, CAPA, verify, close.
CAPA & traceability
Retrain, guardrail, prompt change, eval added, rollback, human-in-the-loop. Verification is the eval that caught it, re-run. The agent is included in Pro and above.
Why Now
Three converging forces
AI entering critical operations — without rigor
AI is being embedded into defense, aerospace, medical, and industrial systems. The DoD Responsible AI Strategy and Implementation Pathway — in force since 2022 — mandates TEVV across every AI program, with more than 60 lines of effort under CDAO oversight. NIST AI RMF is the civilian counterpart. EU AI Act Article 50 applies from August 2026; high-risk obligations were deferred to December 2027. No purpose-built tooling exists to operationalize any of it.
The stand-in tools are not a reliability record
AI GRC captures policy. Observability captures traces. MLSec captures attacks. AI SRE agents close infra incidents. Spreadsheets remain the default binder. None run failure mode → incident → RCA → CAPA → verification on AI behavior. Teams need a cloud-native reliability record, not another MTTR agent.
Just as no manufacturer ships a jet engine without failure analysis
No responsible organization will keep deploying AI into mission-critical environments without structured reliability assurance. That analogy is the translation, not a second product. Hardware reliability is how we already know this works. AIR is the company that makes it the default for AI.
Product
Cloud-native. AI-first. Compliance-ready.
What you can open now
AIR Core Feature workspace: five connected AI systems, a typed failure-mode registry, banded triage, reliability and efficiency views, and corrective actions with verification. Ask the agent about any mode, trend, or dollar figure.
What ships next
Month 5: AIR MVP in production. Year 2: TEVV module, AI system registry, Responsible AI tenet scorecard, red-team findings as a first-class incident. Year 3: standalone agent, traceability matrix, model-card generator, auto-formatted AI Act / DoD AI Review Board filings.
Innovation Pipeline
FRACAS and CAPA for AI — then stronger proof, then control before release.
FRACAS and CAPA for AI
AIR is the system of record for how your AI fails. You classify failures, open corrective actions, and verify that rates and cost actually improved — the FRACAS/CAPA loop, built for AI products.
- Failure Mode Registry
- Auto-Group Repeat Failures
- Reliability and Cost Scores
- Safety-First Priority Ranking
- Corrective Actions with Proof
- AI System Inventory
- Link Failures to Wasted Cost
- Known-Error Playbooks
- Starter Failure Catalog
- Downloadable Reliability Report
For any company running AI in production: find the failure, fix it, prove it worked.
Own the Evidence Stack
Collect only the signals AIR needs — including in locked-down networks — and turn every failure record into proof that maps to major AI rules and auditor packs.
- Self-Hosted Collector
- Map One Record to AI Rules — NIST, ISO 42001, EU AI Act, NAIC
- Auditor Export Packs
- Tamper-Evident Evidence Vault
- Continuous Evidence Collection
- Reliability Budget Alerts
Private networks can adopt AIR. Auditors get packs from real behavior, not checkbox screenshots.
Stop Bad Releases Early
Use the same failure catalog before launch. Simulate risks, block risky releases, and share failure knowledge across your systems and vendors.
- Pre-Release Failure Analysis
- Pre-Release Failure Simulation
- Export Policies to Guardrails
- Block Risky Releases
- Shared Failure Libraries
- Vendor AI Risk Tracking
- Keep Rule Maps Current
Control moves before customers feel the failure — on the same FRACAS spine you already run.
Market
Platform TAM is hundreds of millions. $4.9M is the SOM plan.
DoD TEVV since 2022; EU Art. 50 Aug 2026; high-risk obligations deferred Dec 2027. Category is AI operational reliability / assurance — not whole GRC. Math, citations, and preexisting site claims (including the 38% CAGR) are on Sources.
Business Model
One rate card. Agent included. Capital efficient.
| Tier | ACV | Billing | Notes |
|---|---|---|---|
| Starter | $2,988 | $249/mo card | 5 AI Systems, 30-day trial, no AI features |
| Pro | $6,000 – $7,200 | Annual $6k or $600/mo | 10 AI Systems; agent included; $995 training |
| Enterprise | $35,000 | Annual | 10+ AI Systems, SSO; $10k training; agent included |
| Scale | $125,000 | Annual | Dedicated RDS, priority support; $20k training |
| GOV / Defense | $150,000 | 50% / 50% ATO | On-prem, air-gap, NIST RMF / 800-53 / CMMC / ITAR |
| Add on | ACV | Billing | Notes |
|---|---|---|---|
| Agent Plus | $6,000 | Dual billing | Premium agent capacity on Pro+ |
| Agent Unleashed | $10,000 | Annual | Max agent capacity |
Paying customers by tier — year-end (Y0–Y3)
| Tier | Y0 | Y1 | Y2 | Y3 |
|---|---|---|---|---|
| Starter | 1 | 20 | 78 | 195 |
| Pro | 3 | 29 | 88 | 163 |
| Enterprise | — | 3 | 10 | 26 |
| Scale | — | — | 1 | 5 |
| GOV / Defense | — | — | 1 | 3 |
| Total | 4 | 52 | 178 | 392 |
| Y0 | Y1 | Y2 | Y3 | |
|---|---|---|---|---|
| Recognized revenue | $5.5K | $399K | $1.71M | $3.98M |
| Gross margin | — | 85% | 89% | 91% |
| Operating income | ($388K) | ($477K) | $239K | $1.66M |
| Ending cash | $112K | $689K | $1.65M | $4.47M |
| Exit ARR | — | — | — | $4.91M |
Y0 gross margin is omitted: launch-month COGS sits on almost no recognized revenue. Y3 GM is 90.9%. The $3K/mo marketing plan includes a +30% customer-count uplift versus the no-uplift case — the aggressive assumption in the model. Year-end customer counts are the model’s per-tier mix (AWS customer table) × that 1.30x uplift, with former Agent-standalone logos counted as Pro. Y0 is the live M6–M7 platform MRR (1 Starter + 2 Pro at 1.0x). Bars sum to 4 / 52 / 178 / 392. Agent Plus and Unleashed are add-ons, not extra customers.
Traction
AIR POC live. Self-funded to this point.
- Mar 2026 Idea started — Curtis Bise & Nick Cobb, remote, Texas, self-funded
- Aug 2026 AIR POC live — this mockup, for investor and customer meetings
- M1 $500K pre-seed · three people on payroll from day one (Curtis, Nick, Dev One)
- M5 AIR launches after a four-month build
- M6 AIR GOV / Defense on-prem · NIST RMF · 800-53 · air-gap-capable
- M8 $750K seed on live revenue · ~3 months of AIR in production
- M14–M24 SOC 2 Type II (~$35K) · NIST RMF / TEVV consultant ($50K, M14–M27)
- M24 First profitable month · Y2 operating income positive
- Y2–Y3 TEVV module, tenet scorecard, red-team incidents, then standalone agent and model cards
Team
The rare combination this problem requires
Curtis Bise
Co-Founder & Chief Executive Officer
Technology leader with 11+ years of enterprise-scale software experience. Currently Head of Customer Operations & CRM Technology at Inbenta, leading a globally distributed team of 25+ and reporting directly to the CEO. At GIACT, rose from System Administrator to SVP of Integrated Services — managing 100+ active client integrations, migrating 85% of business operations onto Salesforce, and overseeing a billing system that processed $50M in invoices annually. Named Employee of the Year in 2018 after relaunching the Salesforce organization in under three months. B.A.A.S. in Computer Systems Networking from the University of North Texas. Three Salesforce platform certifications.
Nick Cobb
Co-Founder & Chief Reliability Officer
Reliability engineer with 8+ years supporting defense and aerospace programs across the full system lifecycle. At Raytheon Technologies and Bell Textron, stood up and maintained FRACAS programs across multiple platforms — including airborne targeting systems and military rotorcraft for U.S. Marine Corps programs — facilitated recurring Failure Review Boards, and led multi-disciplinary failure investigations. Performed reliability prediction analyses per defense acquisition requirements, flight qualification testing to MIL-STD-810G, and generated formal reliability deliverables for contract data requirements. Upon returning to Raytheon, stood up FRACAS programs on two active production programs and chaired weekly production Failure Review Boards. B.S. in Aerospace Engineering from Texas A&M University.
Competition
GRC writes documents about AI. AIR writes the record from operations.
AIR is AI reliability engineering — MTBF, a typed failure-mode registry, and traceable evidence — not AI SRE. AI SRE products optimize MTTR with agents that investigate infra incidents; borrow vocabulary where useful, but do not enter that category or invite an MTTR bake-off.
AI SRE closes the incident. AIR closes the loop.
| Capability | SALT AIR | Credo AI | Arize | HiddenLayer | Spreadsheets |
|---|---|---|---|---|---|
| Cloud-native SaaS | Yes | Yes | Yes | Yes | No |
| AI-powered triage | Yes | Limited | Yes | No | No |
| Closed loop (mode → incident → CAPA → verify) | Yes | No | Partial | No | No |
| DoD TEVV / RAI officer tooling | Yes | Policy | No | No | No |
| Evidence from operations, not after-the-fact docs | Yes | Limited | Yes | Runtime | No |
| PLG entry + enterprise expansion | Yes | Top-down | Yes | Top-down | n/a |
| On-prem GOV / air-gap | M6 | Yes | Yes | Partial | n/a |
SALT marks reflect the AIR POC as of August 2026. On-prem GOV is Month 6 on the plan. Commercial deployment begins Year 1.
| Capability | SALT AIRAIR | AI SRE agentsAI SRE | Incident mgmtIncident |
|---|---|---|---|
| Optimizes | MTBF + cost per successful task | MTTR | Coordination / MTTR |
| Failure domain | AI behavior & economics | Infra / services | Infra incidents + comms |
| Persistent typed failure-mode registry | Yes | No (some memory / corpus) | No typed ranked registry today |
| Closed loop (mode → CAPA → verify) | Yes | Partial (remediate incident) | Partial (retros / tickets) |
| Reliability growth / before–after proof | Yes | No | No |
| Evidence for audit / TEVV / program office | Yes | Blameless learning docs | Ops retros, not assurance record |
| Relationship to AIR | — | Feeder / integration | Watchlist if they add typed registry |
Nearest corpus threat: incident-management vendors (PagerDuty, Rootly) already hold structured incident history and CAPA-adjacent workflows. If they add a persistent, typed, criticality-ranked failure-mode registry with trending, they approach FRACAS. Watch Rootly retrospectives and PagerDuty Insights. Resolve / Cleric / Causely remain integration targets, not rivals.
Causely’s persistent causal model is independent validation that structured failure knowledge beats re-deriving from raw signals every time — for infrastructure diagnosis. AIR is the system of record for AI behavior over time (rate, criticality, CAPA, verification), not a real-time causal inference layer.
The Ask
$500K pre-seed round
- Four-month build; AIR launches Month 5. Year 0 is a build year by design, with a thin $5.5K of launch-month revenue.
- Seed at Month 8 on live revenue. AIR GOV on-prem pulled forward to about Month 6 ($150K ACV).
- SOC 2 Type II (~$35K, M14–M24) and NIST RMF / TEVV ($50K, M14–M27) clear enterprise and federal gates.
- First profitable month M24. Y2 operating income +$239K. Y3 +$1.66M. $4.47M cash at M43. No follow-on required in the model.
- Trough cash $112K at Month 7. Bridge protocol: $150–200K F&F/angel standby, arranged before Month 1, drawable only after cost levers.