How The Distillery works
A local proxy between your tools and Anthropic. It distils the context you resend every turn.
Architecture
Your tools
Claude Code, Cline, Cursor, OpenClaw, Hermes
The Distillery
localhost:3080, runs on your machine
Anthropic API
api.anthropic.com
The mechanism
1
Install The Distillery and point your tools at
http://localhost:3080. One environment variable, and everything else stays the same.2
Every request passes through the local proxy. The Distillery distils repeated context and oversized tool outputs before forwarding to Anthropic.
3
Anthropic receives fewer tokens and charges less. Your tools get identical responses, as the distillation is invisible to the AI.
Common questions
Does it slow things down?
No measurable delay for typical sessions. The proxy runs locally and adds sub-millisecond overhead, and the network hop to Anthropic dominates.
Can I audit the code?
Yes. The distillation logic is open source. See the /security/ page for specifics on what runs locally versus what syncs.
What if The Distillery goes down?
Set DISTILLERY_BYPASS=1 and your requests go directly to Anthropic, same as before you installed. You are never locked in.