Independent runtime enforcement — EU-native

Your agents are acting right now.
Can you prove what they did?

Veragent gives you an audit-grade trail and a kill-switch for every AI agent you instrument. Vendor-neutral. EU-hosted. AI Act-ready.

$ pip install veragent try demo →

EU-HOSTED · IRELAND · ZERO LLM SUBPROCESSORS · FREE TIER · 5-MINUTE SETUP

Live audit trail··:··:·· UTC

14:02:11support-l1resolved ticket #4821

14:02:36hr-syncsynced 847 records

14:03:02finance-analyzeranomaly in expense #2847

14:03:29support-l1resolved ticket #4822

14:03:55marketing-gendraft queued for review

14:04:18dev-proxywrite attempt on prod db

14:04:19veragentpolicy P-7 trippedPAUSED

12 agents · trust 94policy P-7 armed

01 — The problem

Agents act at machine speed. Accountability doesn't.

NO INDEPENDENT RECORD

Your agent platform logs its own behavior. When something goes wrong, the vendor is grading its own homework.

OBLIGATIONS ARE YOURS

Under the EU AI Act, the record-keeping and oversight obligations for companies deploying high-risk AI apply from 2 December 2027. The agent vendor doesn’t carry them. You do.

NO KILL-SWITCH

Dashboards show you what happened. They can’t stop what’s happening. Oversight without intervention isn’t oversight.

02 — The mechanism

Three steps to a governed fleet.

Instrument

One install, three lines. Adapters for LangChain, CrewAI, OpenAI Agents, MCP, and Vercel AI.

$ pip install veragent

from veragent import Veragent
va = Veragent(agent="my-agent")
va.track("answered question",
  event_type="tool_call", tool="search")

Set policy

Trust-score floors, permission rules, event thresholds. Rules act for real: deny, escalate, flag, pause, or revoke.

R-7 · when trust_score < 70
effect: deny
remediation: pause

Enforce

Violations are caught at runtime, the agent is stopped, and the intervention itself is written to the audit trail — evidence included.

14:04:18  dev-proxy  write attempt
14:04:19  veragent   R-7 tripped · PAUSED

03 — The instruments

Everything between your agents and the consequences.

LIVE MONITORING

Trust scores and activity for every agent you instrument, streaming in real time.

POLICY ENFORCEMENT

Governance rules with teeth: automatic flag, pause, or revoke when an agent crosses a line.

AUDIT-GRADE TRAIL

Every governed action logged with timestamp, severity, and context. Reject-don’t-truncate ingestion keeps the record honest.

SIGNED ALERTS

Slack, Teams, PagerDuty, and HMAC-signed webhooks the moment something needs a human.

EVIDENCE PACK

Board-ready compliance evidence, generated in one click, exportable to PDF.

TWO SDKS

Python and Node, open source, dependency-free wire client. Five minutes from install to first event.

04 — The wiring

Capture from your stack. Alert to your tools.

Capture

LangChain
CrewAI
OpenAI Agents
MCP
Vercel AI SDK
LangChain.js
any REST client

Alert

Slack
Microsoft Teams
PagerDuty
signed webhooks

05 — The timeline

The deadline moved. The obligation didn't.

The EU's Digital Omnibus — in force since July 2026 — moved the Annex III high-risk obligations, Article 12 logging and retention among them, to 2 December 2027. A deferral, not a repeal: the duties are unchanged and the date is fixed. Veragent's records are compliant by default — append-only, sealed, 180-day retention out of the box on every tier — so readiness costs nothing to have early.

JUL 2026Digital Omnibus in force — Annex III obligations get a fixed date

DEC 2027Annex III obligations apply — Article 12 logging, ≥6-month retention

read what changed →

Pricing