Dev Tools · 1h ago
New tool ContextOS cuts LLM token usage by 97% for code tasks
ContextOS scans a codebase, ranks files by relevance to a task, and exports a context pack using only 15 files instead of thousands. In a test on FastAPI (2,811 files), it reduced token usage from ~284,000 to ~7,998, a 97% savings. The tool runs fully locally, redacts secrets, and can integrate with Claude via MCP.
Meridian48 take
The 97% figure is impressive, but real-world savings depend on task specificity and repo complexity; still, this addresses a genuine pain point for developers using LLMs.
Read the full reporting
I built a tool that cuts Claude/ChatGPT token usage by 97% — here's how it works →
DEV Community
developer-toolsllm-efficiency