Developer zdk released Lowfat, a CLI filter tool that reduces token consumption for AI coding agents by stripping verbose command output. Over 2 months of personal use across 411 command runs, Lowfat reduced 4.4M raw tokens to 360K processed tokens, achieving 91.8% token savings. The tool reached 108 points and 56 comments on Hacker News on June 5, 2026.
Plugin System Enables Custom Filters for Enterprise CLI Tools
Lowfat functions as a lightweight single binary that sits between command execution and AI agent consumption. Rather than feeding agents full verbose output like 'kubectl get -o yaml' or 10,000-line dumps, Lowfat filters out noise and passes through relevant information. The tool features a plugin system that allows users to customize filters per command, adjustable aggressiveness levels, and UNIX-style composable pipes for filter pipelines. It works as an agent hook or shell wrapper and operates locally with no telemetry.
Real-World Usage Statistics Demonstrate Significant Token Reduction
Detailed statistics from 2 months of use show consistent savings across common developer commands:
- kubectl get: 101 runs, 14.4K average raw tokens → 93.9% savings
- grep: 103 runs, 13.5K average raw tokens → 96.2% savings
- docker: 127 runs, 5.5K average raw tokens → 96.1% savings
- git diff: 81 runs, 995 average raw tokens → 57.9% savings
The total across all commands processed 4.4M raw tokens and saved 4.1M tokens, achieving the overall 91.8% reduction rate.
Designed for Customization and Enterprise Use Cases
Unlike bundled solutions, Lowfat keeps its core lightweight while remaining extensible through plugins. The developer emphasizes that the tool is specifically designed for non-public CLI tools and enterprise use cases where teams need custom filters. Users own their data and can create filters for internal or enterprise CLI tools that public solutions cannot access. This customization focus distinguishes Lowfat from pre-built alternatives that may not support organization-specific tooling.
Key Takeaways
- Lowfat reduced 4.4M tokens to 360K over 2 months of use, achieving 91.8% token savings for AI coding agents
- The tool uses a plugin system allowing custom filters per command, with adjustable aggressiveness levels
- Individual command savings ranged from 57.9% (git diff) to 96.2% (grep) across 411 total command runs
- Designed for enterprise use cases where teams need filters for internal CLI tools not supported by public solutions
- Released as open source at github.com/zdk/lowfat with a local-first architecture and no telemetry