Sovereign EU governance · GDPR & EU AI Act

Agents run your business. Your rules run the agents.

Nothing runs until it clears your rules. Every decision, on the record.

Works on your trusted agentic harness
Cursor VS Code Claude Code GitHub Copilot Zed Windsurf Neovim Cody Replit Ollama OpenAI Gemini DeepSeek Grok Perplexity + hundreds more
Runtime governance

We check it before it can do harm.

Every prompt and tool call passes through TapPass before it reaches the LLM or the MCP server. We check it against your policy, then allow it, hold it for a person, or block it, before the action ever executes.

Agent TapPass LLM / MCP
Allowed Held for a human Blocked
🔒app.tappass.ai
How it runs

Two ways to run it.

You choose where enforcement lives. Same rules, same record.

Your harness
enforces here
TapPass
refund > €500 → holdexport PII → ask a human
Backend & tools
Point your harness’s backend URL at TapPass. It sits in the path, checks the rulebook, and passes, holds for a person, or blocks.

TapPass in the path

Your harness connects to TapPass as its backend. Every call runs through it, checked against your rulebook, before it reaches the tool. One URL, nothing to build.

Best when you want a single endpoint your agents connect to, and enforcement handled for you.

one base URLvirtual MCPMCP + models
What you get

The controls, and an AI to set them.

Everything to keep agents in line, with an AI that helps you write the policy. Explore each on the platform.

How it fits

Fits everything your enterprise already runs.

TapPass sits in front of your models, your MCP servers and your gateway, and inside your own app. Scroll to see where it plugs in.

Where it plugs in
Virtual MCPbilling···
create_issueGitHub
post_messageSlack
create_pageNotion
delete_issueLinear
team: financeproject: billing3 of 4 · 12 agents

Three tools, not thirty-four.

Grant tools per team, project or user from any MCP server — or compose a virtual MCP with exactly the tools a job needs.

  • Per-tool permissions
  • A virtual MCP per job
  • Human approval on risky calls
See the MCP gateway →
Modelsgoverned···
GPT-4oOpenAI
ClaudeAnthropic
Mistral LargeEU · France
DeepSeekoff · region rule
via OpenRouterEU-only · BYOK

Every model call, checked.

Sits in front of OpenRouter, LiteLLM or your own router. Whitelist the models you allow; every call is checked before it runs.

  • Whitelist by default
  • BYOK — your keys
  • EU-only models on demand
See the Gateway →
Your API gatewayKong···
→okPOST /refund · €120
→holdrefund €8,200 · over cap
←holdyour gateway holds it
Verdict SDK, inline~200ms

Keep your gateway.

Runs alongside Kong, MuleSoft or Azure APIM — or call the Verdict inline so enforcement stays where your traffic already flows.

  • No rip-and-replace
  • Verdict SDK, inline
  • One rulebook everywhere
See the Verdict →
Every decisionSplunk···
09:41okbilling · refund €120
09:41holdbilling · refund €8,200
09:42blockresearch · export_all
OpenTelemetryredacted in transit

Evidence is a query.

Every decision streams to Splunk, Sentinel or Elastic as OpenTelemetry. Redacted in transit; raw data is never persisted.

  • OpenTelemetry (OTLP)
  • Splunk · Sentinel · Elastic
  • Redacted in transit
See observability →
Request pathhooked···
→okcheck(agent, action)
←holdverdict + reason
→okyour code enforces
Python SDK · PyPIno proxy

Inside your own app.

Instrument your application against the governance server and hook the request path directly — no proxy in between. Published to PyPI.

  • Python SDK on PyPI
  • No proxy required
  • Same rules and record
See the Verdict SDK →
Needs approvalheld···
billing-agent wants to
refund €8,200 to Maria L.
Allow in SlackBlock
Slack · itsmedecided in seconds

Approvals where people already are.

Route holds to Slack, email, or a strong-auth tap with itsme. The right person decides in seconds.

  • Slack, email and inbox
  • Strong-auth with itsme
  • One tap to allow or block
See approvals →
What teams say
“Nothing ships that we didn’t allow.”

We watched first, tuned the rules from what we saw, then switched on enforcement. Live in weeks.

HP
Head of PlatformEuropean fintech
“Our AI pipeline finally feels predictable.”

Every agent action passes a check we own, so our decisions are more confident and our auditors stay calm.

CI
CISOInsurance group
“Auditors got their evidence, automatically.”

Every allow, hold and block lands in our SIEM. No spreadsheets, no month-end scramble.

CL
Compliance leadHealthcare
“We put agents on real refunds in weeks.”

TapPass dropped into the stack we already run. One rulebook, every agent, every model.

HA
Head of AILogistics
“One rulebook, every agent.”

Scoped tools by team and project, held the risky calls for a person. It just works.

VP
VP EngineeringSaaS
How you roll it out

Watch first. Enforce when you’re ready.

Authority is earned, not switched on. TapPass grows from watching to enforcing at your pace.

01 · Connect
Day one

Plug it in

Point your agents at TapPass. One line, nothing changes for them.

02 · Watch
Week 1

See the truth

Every call, every tool, on the record. No rules yet.

03 · Recommend
Week 2

Rules, drafted

TapPass reads the traffic and drafts your rules in plain words.

04 · Enforce
Week 3, on your word

Live

Switch on when you’re ready. From then on, agents stay inside the lines.

Read what matters

Learn from the people governing agentic AI.

Insights and perspectives on running agents safely, from the people building the authority layer.

Ready when you are

See TapPass on your own agents.

Watch first, enforce when you’re ready — live in your EU region in two weeks, on the stack you already run.

No credit card · EU-hosted · works with your MCP, models & gateway
Book a demo