The authority layer for agentic AI. One rulebook, one record, every agent.
Agents are eager. TapPass sits at the boundary between your agents and the models and tools they use, decides every call against rules you own, and keeps a signed record of what happened. On the stack you already run.
✓Free for one person · no credit card
Enforce in the path, or from your own code. Same rules either way.
Pick the entry point that fits each team. They all decide against the same rulebook and write to the same record.
An OpenAI-compatible proxy between your agents and every model and tool. Change the base URL and every call is checked before it runs. No app changes.
OpenAI-compatible~200 msBYOK Policy engine Rules in plain words, decided the same way.The people who own the risk say what worries them. TapPass proposes the rule, they confirm it, and it runs inline on every call, starting in watching mode.
plain languagedeterministicreplay MCP gateway Four tools, not twenty-four.Tools classified by intent, granted per team, project or person, and composed into virtual MCP servers. Risky calls wait for a person.
per-tool permissionsvirtual servers Verdict API You ask. TapPass decides. Your code enforces.Ask before an agent acts and get allow, hold or block with the reason. For teams that keep enforcement in their own code or behind their own API gateway.
allow · hold · blocksigned mandateThree promises hold everywhere. Whichever way a call comes in.
Written once, in plain words, and applied to every agent and framework: LangChain, CrewAI, MCP, Claude Code and your own agents. No rule copied into each team’s code.
Every allow, hold and block from the gateway, the MCP gateway and the Verdict API lands in the same record. Signed, chained to the one before, and yours to export.
A model never makes the call. The verdict comes from your rules and the facts, the same way every time. Replay any decision to prove it, or a draft rule over last week.
- ✓Repoint one base URL, nothing rebuilt
- ✓Redaction and EU routing before a model sees a thing
- ✓Per-tool permissions on every MCP server
- ✓Your traffic keeps its current path
- ✓Fail closed or fail open, per integration
- ✓Mix with the gateway, team by team
Observe. Decide. Record. On every call.
Most of the value will come from agents calling other agents, tools and models across teams and companies. That only works if every call is governed.
Three ways to deploy. One rulebook.
Start in our EU cloud in minutes. Move into your own environment when procurement asks.
We run it for you in a European region. The fastest start, with a DPA on request.
For Enterprise: TapPass runs in your own environment, under your own controls.
Signed images and a signed licence. Air-gapped deployments on request, with our team.
Every OpenAI-compatible model, coding agents, MCP servers, your SIEM and your identity provider.
See what it works with SecuritySix controls on every call, and straight talk on certification.
The security model SovereigntyWhere your data lives, what is kept and who holds the keys.
Data and sovereigntySee it work first. Then decide what it costs.
Which part of TapPass do we need?
Most teams start with the gateway: repoint one base URL and every model and MCP tool call passes your rules. Add the MCP gateway for per-tool permissions, or call the Verdict API where you would rather enforce inside your own code. All of them share the same rules and the same record.
Is a model making the decisions?
No. Your rules compile to deterministic logic. The same facts give the same verdict every time, and you can replay any decision on the record to prove it.
Where does TapPass run?
In the EU, in our cloud. Enterprise customers can run it in their own tenancy or on their own servers.
How long does it take to start?
Minutes for one agent: point the base URL at TapPass and watch. Nothing is blocked until you switch a rule to enforcing.
What does it cost?
Free for one person, with the whole product and no time limit. Team and Enterprise are quoted.
One rulebook for every agent. Start with the one you run today.
Connect one agent and see every call TapPass would have held, stopped or routed elsewhere. Nothing is blocked while you learn.