The AI Gateway

Run your AI workforce like a workforce.

Your people and your agents now work side by side. IRIS8 puts one gateway in front of that workforce — masking what shouldn't leak, holding what needs approval, and staffing routine work with right-sized specialist models instead of frontier prices. Proof at every step.

support-copilot
dev · cursor
claims-agent
SSN 042-68-1174
SSN ***-**-**** · masked
wire $18,500
⤷ held for approval · s.reyes
approved · on the record
request → frontier-lm
routed → specialist-slm  −78%
Masked before it left.50+ data classes, inline, <100 ms
Approved on the record.Every action evidenced automatically
Cost cut. Quality held.Routed to a specialist · 20–40% savings
Trusted by the world's most innovative AI companies
Foundr Uptiq AI North Astra AI ZenMe Vector IntelligenceVector Intelligence Value Buddy Harborline TechHarborline Tech
The problem

Your AI workforce is already here. Nobody’s managing it.

AI is everywhere

Copilots, agents, and AI tools in every team — 60% of it shadow AI that nobody registered, on a surface that grows daily.

Control is inadequate

API keys instead of permissions, email threads instead of approvals — and 35% of enterprises couldn’t shut down a rogue agent.

Proof doesn’t exist

When the examiner asks "why did the AI do that?", logs and attestations aren’t answers. Reconstruction is.

The platform

One gateway. Three products.

Every request through the gateway is secured, governed, and optimized — nothing leaks, nothing rogue, nothing wasted.

See everything. Stop what shouldn’t leave.

One drop-in endpoint lights up your whole estate — live inventory, shadow-AI discovery — then masks sensitive data and blocks injection inline.

  • Five-minute setup — swap a base URL
  • Shadow AI discovered from real traffic, not surveys
  • 50+ data classes masked; findings to your SIEM as OCSF
Explore Secure →
app.iris8.ai/secure/estate
143AI systems
27shadow
$48.2Kspend / mo
commercial-underwriting-v3frontier-lm-7 · lendingActive
support-copilotgpt-class · contact centerActive
unregistered: cursor-agentengineering · 14 seatsShadow

Make every token count.

Semantic caching, right-size routing, and budgets with auto-cutoff — custom logic at the gateway that pays for the platform.

  • 20–40% typical spend reduction
  • Quality-bar routing with automatic fallback
  • Showback that gives finance a bill, not a mystery
Explore Optimize →
app.iris8.ai/optimize
$12,480saved this month
34%cache hit rate
9budgets active
route: gpt-class → small-fastsupport triage · quality ≥ bar−78% cost
cache hit: product FAQ cluster2,114 requests served$0.00
budget stop: growth-experimentsmonthly cap reachedHalted

Answer for every decision. Control every agent.

Per-decision reconstruction and examiner packs today — with approvals, scoped permissions, and the kill switch enforcing on the same gateway in 2027.

  • Any material decision reconstructed in under 15 minutes
  • Examiner packs: SR 26-2 · NAIC · EU AI Act
  • Approvals, action caps, kill switch — enforcing 2027
Explore Govern →
app.iris8.ai/govern/reconstruction/A-58291
Agentcommercial-underwriting-v3.2
ActionRecommend: conditional approval
PolicyCredit Policy §4.2.1 (eff. 2026-03-01)
Sources3 approved · provenance ✓
Modelfrontier-lm-7 · pinned ✓
Sign-offs.reyes · Senior Credit Officer ✓
Packexported · examiner-ready ✓
Specialists, not generalists

Stop paying frontier prices for routine work.

A workforce isn’t all senior generalists — you staff each task with the right specialist. The IRIS8 gateway does the same for AI: every request routes to the smallest model that clears its quality bar, backed by a growing library of specialized, cost-efficient SLMs curated for common enterprise work.

Frontier models where they earn their price. Specialists everywhere else. And because routing happens at the gateway, every decision lands on the record — the savings and the proof come from the same pipe.

Routing & caching · NowSpecialist SLM library · In development
Explore Optimize →
gw.iris8.ai · routing decisions
routesummarize-claims → claims-slm-7b ✓ quality 0.97 −82%
routeclassify-tickets → triage-slm-3b ✓ quality 0.99 −91%
cacherepeat prompt → served from cache −100%
routedraft-credit-memo → frontier-lm-7 bar: frontier
log4 routing decisions → on the record ✓
Why now

AI governance in real-world numbers.

60%

of enterprise AI is shadow AI — undocumented and ungoverned

Industry research
35%

of enterprises admit they couldn't shut down a rogue agent

Writer, enterprise survey
40%

of autonomous agents face demotion or decommissioning by 2027 over governance failures

Gartner, May 2026
<15 min

for IRIS8 to reconstruct any material AI decision, end to end

Our pilot guarantee
Prove, today

When someone asks "why did the AI do that?" — you'll have the answer in minutes.

A registry entry is not an explanation. An audit log is not an argument. IRIS8 reconstructs the decision itself: which policy governed it, which sources grounded it, which model produced it, who approved it.

Explore Govern →

app.iris8.ai/govern/reconstruction/A-58291
Agentcommercial-underwriting-v3.2
ActionRecommend: conditional approval
PolicyCredit Policy §4.2.1 (eff. 2026-03-01)
Sources3 approved · provenance ✓
Modelfrontier-lm-7 · pinned ✓
Sign-offs.reyes · Senior Credit Officer ✓
Packexported · examiner-ready ✓
Integrations

Native integration, zero friction — across the whole AI stack.

Homegrown apps & copilots

Anything speaking an OpenAI-compatible API — swap the base URL and you’re governed.

Agent frameworks & MCP

LangGraph, CrewAI, AutoGen, and MCP servers — SDK hooks and per-tool permissions.

Third-party SaaS AI

Vendor agents and embedded copilots, inventoried and governed like everything else.

Developer endpoints

AI-assisted coding tools and IDE agents — visible through gateway traffic, not surveillance.

For regulated enterprises

Underneath it all: the enterprise AI governance platform for regulated industries.

What starts with one developer pointing an SDK at the gateway scales to what a bank examiner expects — evidence packs, runtime control, and a posture built for SR 26-2, NAIC, and the EU AI Act. Same gateway. Same record. See your industry →

<100 ms

Inline checks at p95 — governance your users never feel.

Split-plane

Hybrid and on-prem/air-gapped gateway for data-sovereign deployments.

OCSF native

Findings flow to the SIEM your SOC already lives in.

$5 / seat

Plus token usage. Free to try — enterprise control when you need it.

The horizon

AI that governs AI — with your policies as its instructions.

At hundreds of agents, human review of every action stops scaling. IRIS8 Guardians are the supervisory layer: always-on agents that watch your operating agents — and file their own evidence.

Explore Guardian Agents →
Reviewers

Evaluate material outputs against policy before they leave the institution.

Monitors

Watch for drift, anomalous tool use, and out-of-role behavior, continuously.

Protectors

Quarantine, revoke, and roll back the moment thresholds are crossed.

Get started

Nothing your AI does is hidden.

Point one app at the gateway today — free — and see your estate light up. Enterprise enforcement is there when you need it.

Start free