Skip to main content

TraceCtrl

TraceCtrl gives security teams and developers complete visibility into every agent action, tool call, and data access — with runtime protection and attack graph risk scoring. Built by CloudsineAI.

What TraceCtrl Does

Agent Visibility

See every LLM call, tool invocation, and agent-to-agent message in real time. Know exactly what your agents are doing.

Security Enrichment

Every span is enriched with security attributes — tool risk categories, input provenance, session correlation, and prompt drift detection.

Topology Mapping

Automatically discovers agent-to-agent and agent-to-tool relationships, building a live topology graph of your agentic system.

Risk Scoring

TAGAAI-powered attack graph analysis scores agents and paths by exploitability — prompt injection, excessive agency, data leakage.

How It Works

TraceCtrl uses OpenTelemetry as the transport layer. You add 3 lines to your agent code, and the SDK captures every operation as enriched spans.
Spans flow through the OTel Collector into ClickHouse, where the TraceCtrl Engine processes them into an agent inventory, topology graph, and risk scores — all queryable via API and visible in the dashboard.

Supported Frameworks

Stable = published to PyPI and tested end-to-end. Source only = the instrumentor package exists in the TraceCtrl repo and can be installed from source (pip install -e ./sdk/tracectrl-instrumentation-<framework>), but is not yet on PyPI. OpenClaw uses its native OTEL export — no TraceCtrl SDK needed. See the OpenClaw instrumentor guide.

Next Steps

Quickstart

Get TraceCtrl running in 5 minutes with Docker Compose.

SDK Installation

Install the SDK and instrument your first agent.