AI agents no longer just answer questions. They make decisions and execute actions on your behalf: they refund, deploy, email and delete, with no human in the moment. TapPass is the layer that decides what each agent may do: gradually, per tool, with every step on the record.
Every connection starts with a dialog like this one. Nobody reads it, everyone allows it. Across your company, agents already hold 0 tools. Their jobs used 0 of them last month.
Allow github-mcp?The agent works, TapPass observes. Nothing is blocked while you learn what it actually does and which rules you're missing.
Sensitive calls stop and wait for a person. One tap to approve or deny, in milliseconds, with the reason spelled out.
Proven policies become deterministic. Not a prompt, not a suggestion: a rule enforced on every call, every time.
You don't adapt to our platform. We understand yours. Write the rule the way your business already says it, "refunds above €500 need my sign-off", and it becomes enforced reality on every agent call.
You should never have to know what an MCP server is. Say which outcomes need control, "issuing refunds", and TapPass covers every technical path that could reach that outcome.
Our two safety principles. Each agent gets the tools its job requires and nothing more, and its authority grows step by step as it proves itself, from watched, to held, to trusted.
🔒
🔒
🔒Every policy compiles to a deterministic check that runs the same way on every call, so the same request always gets the same decision. Enforcement is code you can inspect, not a model's opinion, which makes every outcome repeatable and explainable to an auditor.
Agents act on behalf of your whole company. TapPass gives each role direct control over what agents may do, without tickets and without waiting on each other.
Yours kept working through lunch. The only thing that reached a human was the one call that needed one.
This is the vision: authority for AI lives in one place you own, hosted in Europe under European law. Agents earn authority progressively and hold the least privilege their job allows. You define the rules once, in your language, and the hub enforces them at runtime and syncs them into the platforms your teams already use. Change a rule once and it changes everywhere.
Your rulebook is your data — yours to keep, yours to leave. Export it in the open, any time. Centralise your AI authority in one place without ever being locked in.
OpenAI
Claude
Gemini
Mistral
and every agent framework
Why every agent action must pass through a decision you own, and how runtime enforcement makes agents safe to scale.
Get the free whitepaper →
