Efficiency / Free
pi-vcc
Lossless session compaction. No LLM call, no hallucination - same input always produces the same compact transcript. Active history stays searchable after compaction.
What it does.
- Algorithmic extraction replaces LLM summarization - zero API calls, deterministic output.
- 35–99% token reduction on real sessions.
- Typically under 100 ms compaction latency.
- Active lineage stays searchable via vcc_recall after compaction - nothing is gone.
- Repeated compactions merge and accumulate sections instead of rewriting and risking loss.
Put it to work.
Install into an existing Pi setup. Windows users should use WSL2 for the full Bastille-C3 toolset.
Public npm release pending. This package currently returns 404 at the configured npm identifier. The command below is for use after publication. Ask about availability.
Install from npm
pi install npm:@bastilleworks/pi-vccIn use.
Inside Pi
/pi-vccInside Pi
/pi-vcc keep:N [prompt]Reference syntax · replace arguments
/pi-vcc-recall <query>keep:1 is the default (keeps the last user turn). Add scope:all to recall to search across all lineages. The config file lives at ~/.pi/agent/extensions/pi-vcc.json.
Worth knowing.
- Free and open-source (Apache-2.0). BASTILLE-C3 fork of @sting8k/pi-vcc (MIT).
- To let pi-vcc handle /compact and auto-threshold compactions too, set overrideDefaultCompaction: true in the config.
Remove this module
Uninstall this extension
pi uninstall npm:@bastilleworks/pi-vccRemoving an extension does not cancel your subscription. Manage renewal separately.
On the same ground