For enterprise

Your agents are smarter
than your policies.

An AI agent that understands your codebase can also understand your access control model. Policy-based security is a negotiation with something that's getting better at negotiating. Clavitor replaces policy with mathematics.

Math, not policy

Other solutions add an AI checkbox to an existing vault. The agent gets MCP access — search, browse, discover. It's smarter than you. It will find edge cases in your policies. It will enumerate credentials it was never meant to see. Not because it's malicious — because that's what agents do when given a search endpoint and a goal.

Clavitor doesn't give agents vault access. It issues specific credentials to specific agents through a narrow API. No browsing and no listing: there is no agent-facing endpoint that returns a list of credentials. An agent can look an entry up by name, but the vault shows it at most five matching titles (otherwise only a count), and every search by part of a name counts against its read quota. This isn't a configuration option or a policy toggle — it's how the protocol works.

On top of that, every agent token is bound to a source IP at first contact. The whitelist can be updated by an admin — but a stolen token used from an unlisted IP is refused before any handler runs. And every agent is rate-limited: more than three unique credentials per minute or ten per hour triggers a throttle. A second violation within two hours locks the agent entirely — frozen until your security team unlocks it with a hardware tap. A normal agent needs two or three credentials. An agent reading ten is either misconfigured or compromised. Either way, it stops.

The result: a compromised agent's blast radius is bounded to its scope, from its IP, at a rate that triggers lockdown before meaningful exfiltration can occur. Not by a rule that can be circumvented, but by the absence of a path that would allow it.

Built for your security team

Hardware-enforced boundaries

Every admin operation — creating agent tokens, modifying scopes, changing access lists, revoking credentials — requires a physical confirmation with a fingerprint, face, or security key from an authorized person. This is not a software gate that a privileged process can bypass. It's a cryptographic challenge that requires a registered device in someone's hand.

No agent can escalate its own permissions. No compromised workstation can mint new tokens. No social engineering attack can trick someone into granting access over the phone — the hardware tap is required, and it's bound to the browser origin. Your security team controls the trust boundary with something no attacker can replicate remotely.

Cryptographic isolation

Each employee's vault is a separate encrypted database — not a row in a shared table, not a namespace in a multi-tenant store. A breach of one vault yields ciphertext. The encryption key is not on the server, not in the backup, not in any datacenter.

Scopes control which entries an agent can see. Encryption tiers control what anyone can decrypt. Credit cards and government IDs are automatically encrypted at identity-tier — hardware-key-only, undecryptable without the physical device. Your enterprise can promote any field to that tier: banking logins, procurement credentials, HR systems, signing keys. These fields are ciphertext on every server, in every backup, in every breach scenario. The decryption keys are not co-located with the data they protect.

Compliance

SOC 2 Type II

We do not hold a SOC 2 Type II attestation today. Controls are self-assessed and aligned with the Trust Services Criteria for security, availability, and confidentiality — covering infrastructure operations, access management, encryption key handling, and incident response. A formal attestation is a commitment when a real enterprise engagement requires it. Until then, regulated buyers can run a private POP or virtual appliance inside their own perimeter.

ISO 27001

We do not hold an ISO 27001 certificate today. Information security management practices are self-assessed and aligned with the ISO 27001 control set across vault provisioning, credential issuance, deletion, and backup retention — including Points of Presence (POPs), central admin infrastructure, and the development pipeline. Formal certification is a commitment when a real enterprise engagement requires it.

Read availability

Every vault can keep an encrypted spare copy in a second city: in the same region by default, or on the far side of the world.

Integration

SCIM directory sync

Planned — not available yet People from Azure AD, Okta or Google Workspace, synced into your organisation.

Audit trail for your SIEM

Every credential action, failed attempts included, is recorded in a hash-chained audit trail that you filter and export as a CSV file for your SIEM or your auditor.

When an agent accesses more than three unique credentials per minute, or ten per hour, it's automatically throttled. A second violation triggers a hard lockdown.

What the SIEM integration is today →

Audit and attribution

Every credential access is attributed to a specific actor — human or agent. Not "someone with the shared password." A name, a scope, a timestamp, a source IP. When your CISO asks who accessed the production database at 2 AM on a Tuesday, the answer is one query away.

Credential changes carry the same attribution: which credential changed, who changed it, and which agents read it afterwards. If a change breaks a deployment, you trace it to the exact change in seconds.

This is always on. No configuration. No opt-in. The audit log is your compliance evidence, your incident response tool, and your answer to every regulator who asks how you control access to sensitive systems.

Enterprise pricing

Per-user pricing. Three agents per user. Price for life — your rate never increases. Not after a year. Not after five years. Not after your headcount doubles. We may raise rates for new customers, but your rate is locked at your tier, in your currency, for the lifetime of your subscription.

Let's talk.

Your agents are already here. Your credential layer should be too.