Write it in plain words
“Refunds over €500 need finance sign-off.” “No PII to external tools.” Say it the way you would to a colleague.
plain languagebusiness users
Platform · Policy Engine
Rules in plain words. Enforcement in the request path.
Describe what agents may do in plain business language. TapPass compiles it into deterministic rules that fire the same way on every call, in milliseconds — one source of truth for every agent and every framework.
Your rule, in plain words
“Refunds above €500 need finance approval.”
REFUND > €500→REQUIRE APPROVAL
How it works
Business language in. Enforcement out.
Compiled, not interpreted
Each rule becomes deterministic logic that fires the same way every time — no model in the decision path, no surprises.
deterministicauditable
Fires in milliseconds
The check sits in the request path and adds a few milliseconds. Fast enough for every call, from every agent.
millisecondsevery call
One source of truth
One rulebook for every agent and every framework — LangChain, CrewAI, MCP, your own. Write once; it holds everywhere.
all agentsall frameworks
If it matters to the business
If you worry about it, you can govern it.
Governance is intent-based: rules are written against what an agent is trying to do, not the endpoint it calls — so add or remove an MCP, or change a tool’s definition, and your policy still holds.
Tool governance
Which tools an agent may use, per team, project or user — or compose a virtual MCP with exactly the tools a job needs.
virtual MCPper team
Tool access
Least privilege by default: the tools a job needs, not the thirty-four the server exposes — with human approval on the risky ones.
least privilegeapprovals
Costs
Spend and token budgets, per agent and per team. Hold or block the moment an agent runs past its envelope.
spend limitstoken budgets
Time window
When agents may act. Deploys outside the change window, actions out of hours — held for a person.
change windowout-of-hours
Data protection
PII redacted in transit, EU data residency enforced, bulk exports held for a DPO. Raw data never leaves your region.
PII redactionEU residency
Sequence
The order and steps an agent may follow — govern what it is trying to do, not just the endpoint it called.
multi-stepintent-based
Keep reading
Framework
An AI Agent Risk Assessment Framework for Security Teams
A practical risk assessment framework for evaluating AI agents before production.
12 min read
Least privilege
Give Your Agents Less. On Purpose.
Everyone races to give AI agents more power.
8 min read
Governance
The Governance Gap: Why AI Agent Security Requires a New Category
AI agents deploy faster than governance can follow.
12 min read