wtf is panoptes?

Panoptes is a tool for watching what AI coding agents actually do and keeping a record of it. As companies let agents like Claude Code, OpenAI Codex, Google Antigravity, Cursor, and a project called Hermes Agent run shell commands, read files, and touch production systems, it becomes hard to answer simple questions afterward, such as what an agent did yesterday or whether it touched production credentials. Panoptes solves this by capturing every tool call, file read, and shell command an agent performs and storing it in one searchable log, no matter which agent framework produced it. Each supported agent stores its activity in a different format, so Panoptes normalizes everything into one common event schema and saves it in a local SQLite database. It captures activity in two ways: by running as a proxy that sits between an agent and the internet and inspects its network traffic, or through a native plugin built specifically for the Hermes Agent that hooks directly into its internal events. Once activity is captured, a command line tool lets you check overall statistics, search for specific events, view a full timeline for one session, or trace the evidence behind any single recorded event. Beyond just recording activity, Panoptes includes a policy engine that can actively block risky actions before they happen. Policies are written as YAML files with conditions such as blocking any tool call that touches something tagged as production, and each policy can allow, warn about, or block a matching action. The project is under active development, written in Python, with the core proxy, storage, command line tool, and policy engine already built and tested, while a web dashboard is still planned. Installation works through pip or by cloning the repository from source, and it requires Python 3.10 or newer with very few external dependencies. It is released under the Apache 2.0 license.