Multi-source capture
Collects eBPF system telemetry, MITM API/network traffic, and agent-native JSONL events into a single correlated trace so nothing falls through the gaps between layers.
Deep observability for AI coding agents
Capability requires access. Access creates risk. Mantle closes the gap.
Mantle gives teams a complete, evidence-backed record of what their AI agent did - every tool call, file write, network request, and process spawn - so you can investigate incidents, enforce policy, and ship with confidence.
Main features
Live dashboard
Provides timeline and drilldown views to inspect tool calls, process activity, file changes, and outbound connections in real time.
Taint-based analysis
Tracks how untrusted input moves through your system and flags dangerous paths - like user data reaching a shell command or a sensitive file write.
Why Mantle is needed
AI coding agents are useful, but their runtime behavior is often opaque. Mantle makes that behavior inspectable and testable, so teams can move from assumptions to evidence-backed reviews.
- Reconstruct exactly what happened, step by step, for any agent run
- Surface high-risk operations with the context needed to evaluate them
- Validate agent behavior against policy with repeatable, scriptable checks