Veryl Agent Registry · beta opens September 2026
Every team is building AI agents. Veryl gives each one a verified identity, tests it against your rules before it can act, and writes every action to a record no one can edit. So your whole company can use AI, not just engineers.
A thousand agents or zero — you don't need a fleet to start. Bring one workflow that hurts, and we build and verify your first agent with you in the first week →
Why Veryl exists
Every team is building agents. Almost no one can say which ones are safe.
They pile up in notebooks, repos, and chat threads: unreviewed, unversioned, impossible to tell apart. "We have an agent for that" should be an asset. Without verification, it's a liability with a friendly name.
We're building toward something specific: a company where anyone can put advanced AI to work, because the organization set up trust once and every Veryl product carries it everywhere.
The problem we're solving, with receipts →In SailPoint's 2025 survey of technology leaders, most organizations already run AI agents, and fewer than half have policies governing them. SailPoint sells identity security, so read it as a vendor survey; the gap is still the point.
Among organizations that reported an AI-related breach in IBM's 2025 Cost of a Data Breach study, 97% lacked proper AI access controls. The breached population is a subset, and inside it the pattern is nearly universal.
The same IBM study prices unsanctioned "shadow AI" at an extra $670,000 per breach on average. Governance is cheaper than the incident it prevents.
Sources and the incidents behind them, including the ones we chose not to cite, are on the problem page.
The product
We're building the Registry before anything else, because nothing else is safe without it. It's where your rules get set once, and where every agent proves it deserves to run.
Manage and secure your AI agents. Every agent gets a verified identity called the Agent Passport, earns its verification before it can act, and works inside secure spaces called Rooms with clear limits on what it can and cannot do. Roles keep duties separated, every action lands in a record no one can edit, and one action quarantines a misbehaving agent everywhere at once. Your agents are built, or brought in, and run inside Veryl. That custody is the point: it's what makes fail-closed enforcement possible. If verification can't be checked, the agent waits, and governance bolted on from the outside is advice. This is the product, and everything below shows it working.
The road ahead
Two more products inherit the Registry's rules, in the order that keeps the promise honest.
The trusted source of company knowledge, built by people and agents together, so what agents act on is the same information teams write and rely on. In alpha with a small group of design partners.
Build and launch AI apps on company data without being an engineer, bounded by the access and approvals already set in Veryl Agent Registry. In design; waitlist members shape the beta.
Built for regulated work
Veryl is built for the operations and compliance leaders putting their first agents into regulated work. It verifies agents with real access doing consequential work, inside rules your compliance team can read. This is what that looks like by industry, and each one goes deeper on the industries page.
A pharmacovigilance triage agent that never buries an adverse event. A prior-auth navigator that cites the actual formulary rule. Certification tests the answers against your policies before a patient is ever in the loop.
Use cases →Claims intake that reads PHI inside a scoped Room and nowhere else. A scheduling agent that can see calendars but never charts. Access is set per agent, and the audit log shows every touch.
Use cases →A KYC screening agent whose checks are versioned and re-verified. A suitability reviewer that must cite the rule it applied. When regulation changes, re-verification is the release gate.
Use cases →An NDA redline reviewer that flags what it cannot judge. A discovery classifier with an append-only record of every call it made. Privilege stays scoped, and abstention is a passing grade.
Use cases →A change-order estimator grounded in the actual spec, not a guess. A submittal compliance checker that cites the drawing it checked. Every new version re-verifies before it prices anything.
Use cases →A benefits-guidance agent that must cite the actual rule and say "I do not know" rather than guess. The failure mode becomes a failed verification in testing, not a headline. We wrote up the headline version.
Use cases →Early access
We invite people in order, product by product. Tell us who you are and which products you want.
The platform
A controlled repository with verification at its core and one trust gate behind every surface. Everything in this section is the product opening in September.
Trust · Verification
Agents move through an explicit lifecycle and can't run until they're verified. Every check produces a Verification Report you could hand to your security team.
Illustrative product mock. Names and results are sample data.
Trust · The catalog
The catalog is your org's single source of agents, grouped by trust tier. Every row leads with a status spine, so you never have to read to know whether an agent is safe. You see it.
Illustrative product mock. Agents and counts are sample data.
Trust · Agent Passports
When an agent leaves Veryl, exported to another tool or another company, it carries a Passport: a signed, portable credential of exactly what was verified, when, and by whom.
STATUS: VERIFIED
Governance
Veryl enforces governance in the request path, not in a PDF. Duties are separated by role, every action lands in a tamper-evident log, and the risky moments always pass through a human.
Illustrative product mock. Entries are sample data.
Where people work · Rooms
Rooms are shared spaces where your team collaborates with verified agents in real time. Invite a colleague, add an agent, and work. Every turn is attributed, streamed, and kept.
D. Chen 09:41
Can we cover this specialty request for member 4417 under the new formulary?
Prior-Auth Navigator Verified 09:41
Yes, with step therapy. Policy PA-114 section 3.2 applies. I have drafted the determination and its citation for your approval.
D. Chen 09:44
Citation checks out. Approved, send it.
Illustrative product mock. People and messages are sample data.
The trust system
Color in Veryl means exactly one thing. Blue is action. Green, amber, and red are reserved for trust, never decoration. Learn the code once; read it everywhere.
Cleared for use: everything trusted reads Verified. A filled badge was signed off by a named reviewer; an outlined badge passed automated checks. Both carry a freshness date, and a lapsed badge can't run.
Being checked, or waiting on its builder. In-progress states are never green; there is no such thing as almost trusted.
Quarantined agents are blocked everywhere, immediately. Retired agents remain in the record, struck through, never silently deleted.
The artifact
This is an agent's Passport: the tamper-evident credential behind every badge. Nobody on your team has to read one. The tools they work in check it automatically, every time the agent runs. In plain terms: a sealed inspection sticker your other software can read.
The design and threat model, for your security team →Get started
The Registry beta opens in September 2026. When your invite lands, you start in our managed environment in under 15 minutes, or we provision your own single-tenant deployment on AWS or Azure. Either way, we host and operate it. Nothing for your team to install.
veryl.ai
Veryl Agent Registry opens to beta users in September 2026. Waitlist members get their invites first.