Distill: Open-Source Token Compression Tool Saving 87% on Agent Inputs
arzoo14 · reddit · 2026-08-06
A developer released Distill, an open-source (MIT) tool designed to improve token efficiency for AI coding agents. It consists of an adaptive output-compression skill and an MCP middleware for compressing input context.
Key benchmark findings shared by the author:
- Output compression is limited: Inside Claude Code, output compression is nearly a no-op (+0.3%), and aggressive telegraphic compression actually increased output length (-18%).
- Input-side is where savings happen: By cleaning verbose tool descriptions and noisy logs (ANSI codes, repeated lines) before they reach the model, the tool achieved up to 87% reduction in repetitive logs, soak-tested with 3.95M operations and zero integrity failures.
- Safety mechanism: It includes a structural allowlist hook that verifies destructive/safety-trigger language (e.g., 'irreversible') remains intact, blocking responses if critical warnings are accidentally compressed.
The tool is compatible with Claude Code, Cursor, Windsurf, and other major coding environments.
More from coding & agent
- Developer Shares Claude Skills Tip: Using /wait-what to Quickly Explain Code — mattpocockuk · 2026-08-06
- Opinion: AI Agents Desperately Need a Global Orchestrator and Centralized Approval Page — koltregaskes · 2026-08-06
- Procedurally Generating Joseon Architecture with Claude and Three.js — nptacek · 2026-08-06
- Cloudflare OS Architecture: Lying to AI Agents to Ensure Execution Safety — jedisct1 · 2026-08-06
- DeepSeek API Surprises Developers with Built-in Web Search Integration — max_paperclips · 2026-08-06
- Hard Prompt Intervention: Breaking Claude Out of Coding Doom Loops — teortaxesTex · 2026-08-06