← neuravant.ai
Fleet monitoring
See what your agents are doing, get told when one goes quiet, goes wrong, or changes underneath you —
and stop one immediately when you need to.
What runs today
- Telemetry ingest — agents report events via the Python SDK or a drop-in JavaScript
snippet; events are persisted and drive everything below.
- Pathology detection — detectors watch the event stream for failure patterns
(error spikes, runaway loops, cost anomalies) and raise alerts.
- Quiet-agent watchdog — an agent that stops reporting is itself an alert.
- Model watcher — when the model version behind an agent changes, you are told; a
certified agent on a different model is a different agent.
- Kill switch — database-backed, enforced on traffic routed through our LLM proxy and
the MCP governance gateway.
- Dashboard — agents, alerts and telemetry in your account at
app.neuravant.ai.
Two ways to connect an agent
Python SDK — pip install nail-sdk, then report events from your agent's code.
The SDK is published on PyPI and the telemetry endpoint it talks to is live.
No-code snippet — one <script> tag with a publishable key monitors an
agent-facing page without a build step. Keys are issued in the dashboard.
What this page used to claim
An earlier version of this page rendered a "control tower" against fleet APIs that were never deployed.
We removed it. The dashboard that exists is the one behind login at
app.neuravant.ai; the components listed above are the ones that run,
and /status.html checks them live.
Fleet-wide policy (tool allowlists, egress control, scopes) is applied per tenant at the MCP gateway —
see runtime governance. To onboard a fleet:
team@neuravant.ai.