Skip to main content
Documentation

Ongoing operations

Agent Runtime Telemetry

If you run autonomous AI agents in production behind a runtime governance firewall — an Agent Governance Toolkit, a Sanna-style guardrail, or an in-house equivalent — you can stream its allow, deny, and escalate decisions directly into EuroCompliant. Every event becomes tamper-evident audit evidence, and a denied or escalated action automatically opens a tracked corrective action.

Transcript

If you run autonomous AI agents behind a runtime governance firewall, EuroCompliant can ingest its execution decisions directly. Every allow, deny, or escalate call becomes tamper-evident audit evidence, automatically.

Generate an API key from Integrations, API Keys and MCP, the same key type used for MCP and evidence ingestion. Point your agent runtime firewall's telemetry stream at the agent telemetry endpoint using that key.

As execution decisions arrive, they show up on the system's Agentic Runtime Logs tab: which agent, what action, and a color-coded allow, deny, or escalate decision, each one already written to the tamper-evident audit log.

A denied or escalated decision automatically opens a corrective action. For systems classified High-risk, it is linked straight to Article 14, human oversight, and Article 15, robustness, so the evidence is already mapped to the obligation it supports.

Each event can carry the agent framework's own signed receipt. EuroCompliant stores it verbatim and generates a separate receipt hash of its own, so you can prove exactly what was submitted, independent of whichever signing scheme your runtime firewall uses.

EuroCompliant does not verify that signature, and this does not replace actually implementing human oversight. What it gives you is the other half: proof that oversight decisions were made, at the moment they happened, in a log nobody can quietly edit afterward.

Why this is required

Runtime governance logs usually stay inside the agent framework's own dashboard, disconnected from the rest of your compliance evidence. That makes it hard to show an auditor that human oversight and robustness controls are actually operating, not just documented.

For high-risk AI systems, EU AI Act Article 14 (Human Oversight) and Article 15 (Accuracy, Robustness and Cybersecurity) require operating controls, not policy documents. A denied tool call is exactly the kind of evidence those articles ask for — this feature gets it into your audit trail automatically, without anyone copy-pasting logs.

What EuroCompliant does

POST /api/v1/agents/telemetry accepts one execution decision per call: which agent, what action type (e.g. sql_query, file_write, external_api_call), the decision (allow, deny, or escalate_to_human), which policy fired, and the framework's own signed receipt. Authentication is the same X-API-Key your other integrations already use.

Every accepted event is written to this tenant's hash-chained audit log and returns a EuroCompliant receipt hash you can use to prove what was submitted — separate from, and in addition to, whatever signature your own agent framework attaches.

A deny or escalate_to_human decision automatically opens a corrective action under an "Agent Runtime Governance" compliance plan, and notifies your configured channels (webhook, Slack, or email). If the event is tied to a system currently classified High-risk under the AI Act, the corrective action also links Article 14 and Article 15 as supporting evidence — that link is deliberately not added for systems that aren't high-risk, since those articles don't apply to them.

The system workspace's Agentic Runtime Logs tab shows a live table of recent executions for that system — agent, action, decision badge, policy triggered, and a link to the stored receipt.

Walking through it

1

Generate an API key

Go to Integrations, API Keys and MCP, and generate an API key. This is the same key type used for MCP and the evidence-ingest API.

Open /settings/integrations?tab=api-keys →
2

Point your agent runtime firewall at the ingest endpoint

Configure your agent governance toolkit to POST each execution decision to /api/v1/agents/telemetry with the X-API-Key header set. Include system_id if the event should be tied to a specific registered AI system.

Open /settings/integrations?tab=api-keys →
3

Review runtime logs on the system

Open the relevant AI system and switch to the Agentic Runtime Logs tab to see allow/deny/escalate decisions as they arrive, and open any denied or escalated event's linked corrective action.

Open /systems →

The law

Frequently asked

Does EuroCompliant verify the cryptographic_receipt my agent framework sends?

No. That field is stored verbatim as your framework's own proof of what it decided — EuroCompliant doesn't verify it, since the signing scheme is entirely up to your runtime firewall. EuroCompliant separately generates and returns its own SHA-256 receipt hash for every accepted event, which you can use to prove what was submitted to EuroCompliant specifically.

Is the audit trail this writes to cryptographically signed?

It's a SHA-256 hash chain: each entry embeds the hash of the one before it, so any tampering breaks the chain and is detectable. That's a different (and separately verifiable) guarantee from your agent framework's own per-event signature.

Is agent telemetry ingestion available on all plans?

It requires the API Keys feature, included from the Business plan and above — the same gate as the MCP server and evidence-ingest API.

What happens if system_id is omitted?

The event is still logged as tenant-wide audit evidence, but it won't appear on any system's Agentic Runtime Logs tab and won't trigger Article 14/15 obligation linking, since there's no system to link it to.

Related guides

Try it on your own systems

Everything in this guide runs in the live product. Start a free trial and follow along with your own data.

Start free trial