Govern Behaviour,
Unleash your AI

◊ loz enables AI agents to operate autonomously, safely, and correctly so you can stop reviewing everything and unblock your AI.

◊ loz, a product of Hyperdata Inc., is currently available to select partners. We're excited to discuss agent governance and assess fit. Email us!

Email contact@hyperdata.ca

$ loz build skills/refunds/SKILL.md
◊ translating plain-language policy … 47s

    ✓ parsed       18 rules
    ✓ typed        facts + roles
    ✓ consistent   no contradictions
    ✓ sealed       refunds.v3  #a91f
    ! gaps         2 statements need review

    → review & confirm
      console › drafts/refunds.v3

$ loz gate configure

    ✓ repository   ~/store
    ✓ policy       refunds.v3
    ✓ harnesses    claude code · codex
    ● READY        approvals in console

$ loz decide refund.json

    action   issue_refund
    amount   $240.00
    order    #82014 · age 62d
    policy   refunds.v3

    ✓ within_window       ≤ 90d
    ✓ authorized_role     support
    ✗ requires_approval   amount > $200

    ● DENY   requires_approval
      "refunds over $200 need a manager"
      refunds.v3 §4.2 · decision #d41c

    → what would flip this outcome
      grant   manager_approval
      actor   human · console › grants · sla 4h

agent  ▸ issue_refund(#82014, $240.00)
◊ loz  ◀ DENY  "refunds over $200 need a manager"

agent  $ loz gate status

    policy   refunds.v3
    tool     issue_refund
    ✗ requires_approval   amount > $200

    → unblock   grant manager_approval
      approve   console › grants
      retry     original call resumes

agent  ▸ waiting on approval …

◊ loz  ◀ ALLOW  grant consumed · decision #e1b9
agent  ✓ issue_refund(#82014, $240.00)

Grants 1 pending dana@store.com

PENDING 16m refunds.v3

Issue a $240.00 refund for order #82014

issue_refund → #82014 → $240.00 support-agent

Why it is blocked

“Refunds over $200 need a manager’s approval.”

refunds.v3 §4.2 · original proof

Relevant to this decision

Refund amount over the $200 approval limit$240.00

Order age within the 90-day window62d

Granting will assert

manager_approval(#82014) human-attested

Approval resumes the blocked call.

Write Policy - Correctly

Your plain-language policies - business rules, authorization, and behaviour - verified for logical correctness, versioned, and composable.

Govern Behaviour - Autonomously

Deploy contextually aware controls for behaviour, access, and tools that balance autonomy with complex policy compliance and auditing.

Who is this for?

Teams using and creating coding agents, AI workflows, and chatbots - where business rules, authorization, and tool-use policy are important.

How do I use it?

Build ◊ loz policy artifacts from your existing prompts, skills, and policies, or create new ones on the fly. Deploy them to your agent fleet to enforce, observe, audit, and iterate.

Organizations

We support multidisciplinary teams with WorkOS-powered auth, web-based policy management and auditing, and comprehensive developer tooling.

Developers

Use our API, CLI, web app, and SDK with integrations for popular harnesses, control planes, and agent and guardrail frameworks.