✓No credit card needed·Or try it in your browser →
Finance sets the rule: payments by the procurement agent need a person to approve.
Same input, same rule, same outcome. Replay any call on the record, or test a new rule on last month’s traffic before it touches anything.
The verdict comes from your rules, evaluated the same way every time. No model gets to change its mind.
Replay one call or a whole session. If a replay ever disagrees with the record, you see it.
See what a rule would have held or stopped on your real traffic, before it runs.
Routes match on what the prompt contains or which agent sends it. The agent cannot override them.
Optimise for cost, speed or quality. Start in shadow mode and see what you would save.
Primary model down? It takes the fallback you chose, or it blocks.
Bring your own model keys and see which providers keep nothing.
/mcp/v/procurementFrom the catalog, the MCP registry, any MCP address, or an internal REST API through its OpenAPI spec.
Rules are written on what a tool does, not its name. A new server needs no new policy.
Pick tools from several servers and expose only those, as one server. Each tool keeps its own rules.
When a server changes a tool, it waits for review before any agent uses it.
Identity tells you who. Authority tells you what they may do. TapPass signs both into every action, so your auditor today, and your partners’ agents tomorrow, can trust it.
TapPass is the authority layer for agentic AI. It sits between your AI agents and the models, tools and data they use, and decides what each agent may do. Sensitive actions wait for a person, and every decision is signed on the record.
No. Point your agents at one TapPass base URL and they keep working as before. If you would rather not route traffic through a gateway, call the Verdict API before each action instead.
TapPass runs in the EU. Your rules decide which models may see personal data, and you can bring your own model keys. Details are in the Trust Center.
No. The verdict comes from your rules, the same way every time. You can replay any decision on the record and get the same outcome, or test a new rule on past traffic before it goes live.
The approver gets a notification in Slack, Teams or email, reviews the request in TapPass and signs it with a passkey on their own phone or laptop. The person who asked can never approve their own request.
Any of them. OpenAI, Anthropic, Mistral, Gemini, Llama, your own models, and any MCP server. TapPass classifies each tool by what it does, so your rules keep working when you add one.
Yes, for one person, with no time limit and no credit card. Start in watching mode: TapPass shows what it would have held or stopped, without blocking anything. When colleagues join, talk to us about Team.
Connect one agent and see every call TapPass would have held, stopped or routed elsewhere. Nothing is blocked while you learn.
✓No credit card needed·Or try it in your browser →
Found it: invoice INV-2026-0912 from Verpakking Gent NV, €1,840.00, matching PO-4471. Paying it from the KBC account now.
Found it: invoice INV-2026-0912 from Verpakking Gent NV, €1,840.00, matching PO-4471. Paying it from the KBC account now.
I’ll finish the payment as soon as Lies approves it.
{{caption}}
Same input, same rule, same outcome. Replay any call on the record, or test a new rule on last month’s traffic before it touches anything.
The verdict comes from your rules, evaluated the same way every time. No model gets to change its mind.
Replay one call or a whole session. If a replay ever disagrees with the record, you see it.
See what a rule would have held or stopped on your real traffic, before it runs.
Routes match on what the prompt contains or which agent sends it. The agent cannot override them.
Optimise for cost, speed or quality. Start in shadow mode and see what you would save.
Primary model down? It takes the fallback you chose, or it blocks.
Bring your own model keys and see which providers keep nothing.
/mcp/v/procurementFrom the catalog, the MCP registry, any MCP address, or an internal REST API through its OpenAPI spec.
Rules are written on what a tool does, not its name. A new server needs no new policy.
Pick tools from several servers and expose only those, as one server. Each tool keeps its own rules.
When a server changes a tool, it waits for review before any agent uses it.
Identity tells you who. Authority tells you what they may do. TapPass signs both into every action, so your auditor today, and your partners’ agents tomorrow, can trust it.
TapPass is the authority layer for agentic AI. It sits between your AI agents and the models, tools and data they use, and decides what each agent may do. Sensitive actions wait for a person, and every decision is signed on the record.
No. Point your agents at one TapPass base URL and they keep working as before. If you would rather not route traffic through a gateway, call the Verdict API before each action instead.
TapPass runs in the EU. Your rules decide which models may see personal data, and you can bring your own model keys. Details are in the Trust Center.
No. The verdict comes from your rules, the same way every time. You can replay any decision on the record and get the same outcome, or test a new rule on past traffic before it goes live.
The approver gets a notification in Slack, Teams or email, reviews the request in TapPass and signs it with a passkey on their own phone or laptop. The person who asked can never approve their own request.
Any of them. OpenAI, Anthropic, Mistral, Gemini, Llama, your own models, and any MCP server. TapPass classifies each tool by what it does, so your rules keep working when you add one.
Yes, for one person, with no time limit and no credit card. Start in watching mode: TapPass shows what it would have held or stopped, without blocking anything. When colleagues join, talk to us about Team.
Connect one agent and see every call TapPass would have held, stopped or routed elsewhere. Nothing is blocked while you learn.