XDOG

Tools for building AI agents and managing LLM deployments.

GitHub →

xdog / Packages / eyes / Overview

Independent observations, explicit JSON destinations.

Eyes collects file/time evidence, turns it into events and routes it to explicit handlers. It never owns a model/tool loop or decides whether a user received an Agent's response. The CLI and installed model tool manage the same user-local shared daemon.

Highlights

  • Versioned JSON with file, one-off at, interval and zoned cron sources.
  • Frozen definitions, validation, detectors and explicit handler routes.
  • Durable owner-scoped registrations, source cursors and delivery checkpoints.
  • One daemon under ~/.local/share/xdog/eyes, honoring XDG_DATA_HOME.
  • Application-scoped eyes tool and discoverable skill; no default body injection.
  • Unknown/interrupted effects are not automatically replayed.

Start with a definition

pip install xdog-eyes
xdog-eyes validate /absolute/path/to/eyes.json
xdog-eyes --owner repo-review start /absolute/path/to/eyes.json
xdog-eyes --owner repo-review list

First copy/edit a shipped example with a real receiver, existing destination and deliberately scoped source roots. The placeholder path above is not an observation created for you.

See Reference for controls and Skills for Goal's optional task observation.