docs

Pheebs docs

What Pheebs is, what it stores, and how to run it.

Pheebs sits inside your AI coding agent on a hook. Every time something happens in a session (a prompt goes in, a tool runs, a model changes, the context compacts) it writes one line of JSON describing the shape of that moment, and nothing about its contents.

It exists because nobody has settled what "using AI well" actually looks like. Adoption numbers are easy and say almost nothing. Pheebs is measures what is really happening inside AI sessions, how often, and in what patterns.

Quickstart

Install from npm, then register the hooks:

# install
npm install -g pheebs

# interactive: pick your agents, then hooks and OpenTelemetry are registered
pheebs init

# check the wiring
pheebs doctor

That is the whole setup. Pheebs fires in the background during your sessions and nothing changes about how you work. Read your own data at any time. It is plain text:

tail -qn 20 "$(ls -t ~/.pheebs/logs/*.jsonl | head -1)"

Where to go next

Which agents work

AgentStatusHook entries
Claude Code Fully supported15 entries across 14 hook types
Cursor Supported13 entries across 13 hook types
Codex Fully supported10 entries across 10 hook types

Fully supported means hooks plus OpenTelemetry. Supported means hooks only. All three share the same log format and the same remote transport. The ai_tool field identifies the source.

Requirements

  • Node.js 18 or newer.
  • A git repository, for automatic codebase and developer detection.
  • Claude Code, Cursor, or Codex with hooks support.