The road ahead
Where this goes.
Veryl Agent Registry ships first; its beta opens September 2026. This page is the rest of the plan, and we label it what it is: direction, not shipped product. Stage pills and mock labels tell you what exists today.
The vision
Five keys. One architecture.
Everything we ship serves one of five keys. So does everything on the road ahead.
Trust
Every agent carries evidence of what it is, what it can do, and whether it still deserves to run. A person reads it at a glance; a machine reads it over an API. And it stays alive: verification expires, renews, and travels with the agent as a signed Passport.
Governance
Control that holds at ten agents or ten thousand. Duties are separated by role. Every action lands in an append-only record. The risky moments wait for a human. And quarantine pulls a bad agent out of everything, everywhere, at once.
Where people work
Agents go where people already are: Teams and M365 Copilot, Claude and ChatGPT, email, phone, SMS, the API, and every new door as it opens. Each door leads to the same gate, so trust follows the agent through all of them.
Tools for agents
A trusted agent that can't do anything is a chatbot. Veryl agents get capability on rails: tools and connectors, grounded knowledge, memory, durable workflows with human gates, even delegation to other verified agents. All of it within strict caps.
The marketplace
When trust is portable, agents become an economy. The best agent for a job gets found, licensed, and put to work in minutes, whether it was built down the hall or by a team you've never met. The Passport does the vouching.
On the way
Planned for the Registry.
Two capabilities we're building toward, shown as design mocks so you can react to them before they exist.
Where people work · Every door
Your agents, in the tools your team already uses.
An agent locked in a dashboard doesn't get used. Veryl agents install into the tools your company already runs on, and trust travels with them: quarantine an agent, and every surface returns the block notice on its next message.
- Microsoft shops. One generated app package installs an agent into Teams and M365 Copilot; the bot identity, relay, and manifest are handled for you.
- AI assistants. Claude, ChatGPT, and Gemini connect through one secure door. Each person grants their own access, can lose it individually, and every question is attributed to whoever asked it.
- A real mailbox, safely. An agent can hold delegated access to Gmail, Outlook, or nearly any IMAP mailbox. Reads and drafts are audited; every send is approved by a human first; deletes are never permanent.
Illustrative product mock. Sample data.
Tools for agents · Workflows
Agents with real capability, on rails.
A trusted agent still has to do the work. Veryl gives agents tools, knowledge, and memory, then chains them into durable workflows where the risky steps wait for a human.
- Tools and connectors. Managed connections to calendars, mailboxes, and your systems, granted per-agent and revoked per-agent.
- Knowledge and memory. Grounded retrieval with PII redacted at ingest, plus org-scoped memory of past interactions: scrubbed summaries, never raw transcripts, always labeled apart from verified knowledge.
- Durable workflows. Multi-step, resumable runs with approval gates, scheduled triggers, and parallel steps that survive restarts.
- Agents that delegate. Every user's built-in Executive Assistant can hand a sub-task to any verified agent, bounded by depth, fan-out, and cost caps.
Illustrative product mock. Steps are sample data.
The expansion
Verification for agents you don't host.
Today, enforcement requires custody: agents run inside Veryl so a block is a block. The next tier extends the trust layer beyond our runtime: we run verification against an agent hosted anywhere, issue it an Agent Passport, and expose its live status. We can't quarantine an agent we don't run, but revoking its Passport tells every relying party to stop trusting it, and the public Passport specification exists so they can check that for themselves. Staged after the Registry.
And everything around it
A complete home, not a feature.
Workflows
Chain agents into durable, resumable runs, with human approval gates, scheduled triggers, and parallel steps that survive restarts.
Veryl Canon
The trusted source of company knowledge, used by people and agents alike. Personal information is redacted when documents come in, with retention you control.
Connectors & channels
Bring agents to where work happens: inbound channels for email and chat, outbound webhooks, and managed connectors to your tools.
Executive Assistant
Every user gets a built-in assistant that answers product questions, runs guided demos, and delegates work to your org's verified agents, within strict depth and cost caps.
Delegated mail
Agents can work a user's mailbox: Gmail, Outlook, or nearly any IMAP. Every send is human-approved, every action audited, and trash is always reversible.
Agent memory
Agents remember prior interactions across surfaces as scrubbed one-line summaries, never raw transcripts, kept clearly apart from verified knowledge.
Help decide what ships next.
Waitlist members shape the beta, and design partners shape the roadmap. Tell us which of these you need first.