Put your agents behind a guard. Start with a four-week pilot.

VITNA is a containment layer for AI agents. In guard mode it sits in front of the stdio MCP servers you put behind it. Calls to the tools you mark as risky wait for a person, and if nobody answers in time, they don't run. No API key can decide a hold, including the agent's own. Claimed accounts get Ed25519-signed records anyone can verify offline with our published key.

The pilot puts the guard on one to three of your agent workflows in staging, with your people deciding real holds. It ends with a findings report and a signed evidence package. Fixed scope. Fixed price.

Security and platform teams already running AI agents against real systems. And companies that build agents and sell them to enterprises.

Your compliance team keeps its job. VITNA gives them signed records of what was held, what was decided, and when.

Four weeks, fixed scope.

Half on signing, half on delivery of the findings report. If you continue to an annual plan, the pilot fee is credited to your first year.

Before the pilot. A 30-minute scoping call. You describe your agents and what they reach. If it doesn't fit, I say so.

Week 1. An inventory of your agents and the tools each one reaches. Then a hold-policy session: which calls wait for a person, who approves them, and how long the window stays open (from 30 seconds to an hour). A held call nobody decides in time is blocked.

Week 2. VITNA's guard goes on one to three agent workflows in your staging environment. Holds and blocks go to your alert webhook as a JSON POST; a Slack incoming webhook works as it is. Bait and canary holds are on: decoy tools and planted values that no real task should touch, and practice holds that show whether someone is answering.

Week 3. Your people decide real holds. We tune the policy from what they see.

Week 4. A findings report: what was held or blocked, what sits outside the guard, and what to wrap next. A signed evidence package you can verify offline. A walkthrough call.

What you provide

• A staging environment where your agents call stdio MCP servers.
• An engineer for a few hours a week.
• The people who will approve holds.

The pilot's statement of work and our data processing agreement are sent to you before signing.

Then an annual plan, if the pilot earns it.

No published price. It is quoted after the pilot, from what the pilot found.

1. Team features at higher limits. 6,000 ingest requests a minute instead of Team's 2,000, and no cap on active engagements instead of Team's 500.
2. Signed evidence exports for audits. The evidence package and the signed bundle a session saves when it closes, holds included, verifiable offline with our published key.
3. CEF and LEEF export for your SIEM. Events export in both formats from GET /api/cef.
4. A named contact. Collin Obey, the founder, with a response target in business hours, Saskatchewan time.
5. A quarterly policy review. What was held, what was decided, and what the policy should change.
6. A say in what's built next. Your requests are weighed when the roadmap is set.

What this isn't.

• Not a sandbox or a penetration test. The guard sees only the stdio MCP servers it wraps. Anything an agent reaches another way is outside it, and the findings report names what is.
• No SOC 2 report yet. We answer your security questionnaire in writing, and both sub-processor registers are published: the VITNA product and engagements.
• No uptime commitment and no 24/7 support. Support comes from the founder, in business hours.
• No SSO or approver roles yet. Everyone who decides holds signs in to the same VITNA account, so a decision record names the account and how the hold was decided, not the individual.
• A session's signed bundle can go missing. Some clients stop the guard before it saves; VITNA's decisions on each call still come back in the signed evidence package, without what happened to each hold.
• Not legal advice, and not a certification. VITNA records what was held and decided. What that means for your obligations is your counsel's call.

• Try the demo A labelled simulation in your browser: a scripted agent on pretend files reaches for risky tools, and you decide its held calls.
• Verify a signed evidence package yourself Offline, with our published key.
• What VITNA sends and stores Exactly what leaves your machine, and what is kept.

Thirty minutes to find out whether this fits.

Not legal advice. Not a certification. VITNA keeps the record; your counsel and your auditors decide what it means.