{% macro saver_toggle(setting_key, enabled) %} {% endmacro %} {% macro saver_savings(class_name) %} {% set s = saver_stats.get(class_name) if saver_stats else None %} {% if s %}

saved {{ "%.1f"|format(s.saved_kb) }} KB across {{ s.requests }} request{{ "s" if s.requests != 1 else "" }} ({{ "%.1f"|format(s.avg_pct) }}% avg) — since restart

{% endif %} {% endmacro %}

RTK

Recommended

Recursive Token Kompression — compresses conversation history to reduce input tokens. No quality loss for long conversations.

{{ saver_savings("RTKSaver") }}
{{ saver_toggle("saver_rtk_enabled", settings.saver_rtk_enabled == 'true') }}

Caveman

Injects a brevity system prompt so responses skip filler and pleasantries. Aggressive token savings, may affect nuance. Choose brevity level:

{% set cl = settings.saver_caveman_level %} {% for level in ['lite', 'full', 'ultra'] %} {% endfor %}
{{ saver_savings("CavemanSaver") }}
{{ saver_toggle("saver_caveman_enabled", settings.saver_caveman_enabled == 'true') }}

Headroom

External

Routes the conversation through a locally running Headroom compression proxy (pip install "headroom-ai[proxy]" && headroom proxy --port 8787). Fails open — if Headroom is down, requests pass through unchanged.

{{ saver_savings("HeadroomSaver") }}
{{ saver_toggle("saver_headroom_enabled", settings.saver_headroom_enabled == 'true') }}

Ponytail

Summarizes older messages into a compact summary. Choose compression level:

{% set pl = settings.saver_ponytail_level %} {% for level in ['lite', 'full', 'ultra'] %} {% endfor %}
{{ saver_savings("PonytailSaver") }}
{{ saver_toggle("saver_ponytail_enabled", settings.saver_ponytail_enabled == 'true') }}