AgentSight – eBPF observability for AI agents, no code changes
Source
github.com
Author
forrestly
Date
Context
AgentSight, documented in Alibaba's ANOLISA project, observes AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → activity from the operating system rather than requiring logging code inside each agent. On Linux, its eBPF probes capture model-call traffic, process events and file operations, which support usage analysis and investigation of interrupted sessions. User guide.
That placement changes the deployment requirements. The probes need Linux kernel support and root or CAP_BPF privileges. The macOS tools instead collect existing session files and display them in a dashboard; they do not provide the Linux kernel-level capture. Avoiding agent-code changes still requires installing and operating the observer.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.