Platform · Policy engine

Rules in plain words. Decided the same way, every time.

Say what agents may do the way you would tell a colleague. TapPass turns it into deterministic rules that run inline on every call, for every agent and every framework.

✓No credit card needed

Finance writes“Payments by the procurement agent over €1,000 need a person to approve.”
TapPass compilesRULE PUR-1 · Purchase approvals
WHEN agent = procurement-agent
AND intent = move money
AND amount > 1,000 EUR
THEN hold for an approver
Next callProcurement agent pays €1,840.00 to Verpakking Gent NVHeld for Lies PeetersSame input, same answer. Every time.
01What you can govern

If you worry about it, you can govern it.

Rules follow what an agent is trying to do, not the endpoint it calls. Add an MCP server or change a tool, and your rules still hold.

You say“The procurement agent only uses the tools procurement needs.”
TapPass runsGRANT procurement-agent tools = virtual server "procurement" DENY everything else
What happensProcurement agent calls zendesk.delete_ticketBlocked · not in its job
02Watch, then enforce

See what a rule would do. Before it does it.

WatchingRules record, never act

Every call that would have been held, stopped or redacted is on the record. Nothing is blocked while you learn what your agents really do.

EnforcingRules hold and stop for real

Switch a rule on when you trust it. The same rule, now deciding, with a person signing what matters.

Test before you switch it on
Draft ruleHold payments to a new beneficiary
Replayed over the last 30 days · 128 payments
Would pass125
Would hold3
Would block0
Read only. Nothing changes until you switch it on.
Replay a decision
10:07Procurement agentPay €1,840.00 to Verpakking Gent NVHeld · approved
Replay the decision Replaying under the policy in force… ✓ Same decision. Held for a person, under PUR-1.
On the recordHold · ask a person
Replayed nowHold · ask a person
03In the product

One rulebook. Every agent, every framework.

Write a rule once and it holds for LangChain, CrewAI, MCP, Claude Code and your own agents, through the gateway or the Verdict API.

TapPass concerns: the rules for Vandaele Logistics, grouped by what the business worries about
Next: the MCP gatewayFour tools, not twenty-four. Per agent, per job.
Questions

The policy engine, in plain answers.

Something else on your mind? Talk to our team.

Is a model making the decision?

No. Your plain-language rule is compiled into deterministic logic. The verdict comes from that logic, the same way every time, and you can replay any decision to prove it.

Who writes the rules?

The people who own the risk. Finance, security or the DPO describe what worries them in plain words; TapPass proposes the rule, and they confirm it.

What happens when we add a new MCP server?

Nothing breaks. Rules are written against what an agent tries to do, such as paying, deleting or sending data outside, not against specific endpoints. New tools are classified and your rules apply.

Can we try a rule without risk?

Yes. Every rule starts in watching mode, where it records what it would have done without acting. You can also replay a draft rule over past traffic before switching it on.

Does it work with our frameworks?

One rulebook covers LangChain, CrewAI, MCP, Claude Code and your own agents, through the gateway or the Verdict API.

Say what worries you. Watch the rules work, for free.

Connect one agent and see every call TapPass would have held, stopped or routed elsewhere. Nothing is blocked while you learn.

✓ Free for one person✓ No credit cardBacked by Start it @KBCVLAIO✓ EU hosted✓ Built in Belgium✓ Your own model keys✓ Encrypted at rest and in transit✓ DPA on request