Discover and explore top open-source AI tools and projects—updated daily.
eunomia-bpfAI agent observability via eBPF
Top 94.8% on SourcePulse
Summary AgentSight offers zero-instrumentation observability for LLM agents, monitoring system-level interactions via eBPF. It captures encrypted traffic and subprocesses without code changes, providing deep insights into AI agent behavior with minimal performance overhead. Ideal for engineers and researchers needing comprehensive, non-intrusive monitoring.
How It Works Leveraging eBPF at the kernel level, AgentSight intercepts SSL/TLS traffic and monitors process events, bypassing application instrumentation. Data streams through a Rust analysis framework and is visualized via a React/TypeScript frontend. This system-level approach captures critical interactions like subprocesses, raw encrypted payloads, and file operations, often missed by application-level tools.
Quick Start & Requirements Installation via Docker or source build.
docker run --privileged --pid=host --network=host -v /sys:/sys:ro -v /usr:/usr:ro -v /lib:/lib:ro ghcr.io/eunomia-bpf/agentsight:latest record --comm pythongit clone --recursive ...), install deps (make install), build (make build).http://127.0.0.1:7395.Highlighted Details
--binary-path for statically linked SSL libraries.Maintenance & Community
Project welcomes contributions; links to design docs provided (CLAUDE.md, collector/DESIGN.md, docs/why.md). Specific community channels or maintainer details are absent from the README.
Licensing & Compatibility MIT License, permitting commercial use and integration into closed-source projects.
Limitations & Caveats
Requires Linux (kernel 4.1+) and root. Applications statically linking SSL libraries need --binary-path for accurate SSL capture. Standalone tools (browsertrace, stdiocap) exist for niche use cases.
1 week ago
Inactive