{% extends "_base.html" %} {% from "_partials.html" import avatar, number %} {% set active = 'savings' %} {% block title %}trovex — savings{% endblock %} {% block topbar_meta %} {{ "{:,}".format(totals.saved) }} tokens saved · last {{ days }}d {% endblock %} {% block head %} {% endblock %} {% block content %}
Savings accrue as your coding agents call trovex for docs.
Once queries land, this dashboard shows the modeled tokens saved versus reading
the top-3 .md files by hand.
| avatar | user | queries | would have read | saved | ratio |
|---|---|---|---|---|---|
| {{ avatar(u.user, 22) }} | {{ u.user }} | {{ u.queries }} | {{ "{:,}".format(u.would_have_read) }} | {{ "{:,}".format(u.saved) }} | {{ "%.0f"|format(u.ratio * 100) }}% |
Without trovex, an agent typically does Glob or
Grep to find matching .md files, then
Read the top 2-3 candidates to figure out which one is
canonical. With trovex, they get a ranked list with status
markers and read just the top-1.
would_have_read = ∑ tokens_est(result[0..3]) actual_read = tokens_est(result[0]) overhead = tokens(trovex output) saved = would_have_read − actual_read − overhead
Conservative — assumes the agent does still read the top result. If markers reveal that all results are stale or duplicates, real savings are higher and not captured here. These are modeled estimates over real query logs, never guaranteed exact.