Requests and model calls
Connect prompts, model and version, response identifiers, and policy decisions into one execution trail.
AI Agent Audit Evidence
Build an audit trail for AI agent decisions and tool calls
An AI response does not show which tools ran, who approved an action, or what changed in an external system. The Rantropy package connects requests, model activity, tool calls, approvals, and external effects in an audit trail that the institution can retain and review.
Released for public procurement
This is a separate product from the randomness API and can be purchased through KONEPS and Venture Nara. Deployment scope is agreed against the institution's environment and security policies.
Evidence scope
Record the execution path, not only the answer
The evidence model connects who requested what, which model and tools ran, and what happened next.
Tool calls and external actions
Record tool activity that affects real systems, such as search, data changes, and message delivery.
Approvals and exceptions
Include human approvals, rejections, retries, and overrides so responsibility boundaries remain visible.
Deployment direction
Deploy the package for the institution's environment
Deployment scope and operating policy are defined against institutional requirements for data movement, retention, access control, and independent verification.
Package integration
Connect evidence collection and review capabilities to existing AI applications and agents.
Security and retention policy
Align restricted-network operation, account permissions, encryption, and retention with institutional policy.
Independent verification
Prepare tools that can inspect exported evidence without access to the originating service.
Evidence lifecycle
Keep the schema, retention, and export rules reviewable
The package defines how event versions evolve, how protected fields are handled, when each copy expires, and what an independent review package contains.
Versioned audit schema
Use stable execution identities and typed events for model calls, policy decisions, tool attempts, approvals, and external outcomes.
Retention and deletion
Apply documented lifecycle rules to active records, indexes, exports, backups, and recoverable copies.
Self-contained exports
Deliver a machine-readable manifest with schemas, checksums, integrity links, scope, and recorded exclusions.
Technical guides
Define the evidence boundary before deployment
Review the events, identities, integrity controls, and failure paths that make an AI audit record useful outside the originating application.
What an AI agent audit trail must recordConnect authority, model activity, tools, approvals, retries, and external effects under one execution identity.Read guideTurn AI tool-call logs into reviewable evidenceDesign canonical events, privacy controls, integrity checks, and failure tests for tool calls.Read guideDesign a versioned audit log schema for AI agentsDefine the common envelope, event types, compatibility rules, and privacy fields needed for reliable review.Read guidePlan retention and export for AI audit evidenceMap every recoverable copy, retention rule, export manifest, and independent verification step.Read guide
AI evidence starts with the real workflow
Tell us which models and tools you use, where approvals happen, and what an auditor must be able to verify. We will help define the events that matter.