172
AI • DevTool • Productivity
Coding agents (Cursor, Windsurf, Cline, Aider) resend thousands of tokens every turn — including redundant comments, ANSI logs, whitespace, and repetitive file diffs. You end up paying full input token pricing on noise the LLM doesn't need.
CntxTrim is a high-speed reverse proxy that cleans and optimizes prompt payloads in-flight before they reach your AI provider, cutting token spend by 30% to 60% with sub-5ms overhead.
Key Highlights
30%–60% Token Savings: Intelligently prunes code comments, whitespace bloat, and noisy terminal traces while preserving 100% of functional code logic and reasoning instructions.
Sub-5ms Latency: Engineered for real-time developer workflows so your completions and agent tools feel just as instant.
Provider Cache Friendly: Formats prompt prefixes to maximize upstream provider prompt caching discounts.
Universal BYOK (Bring Your Own Key): Works transparently with your existing Anthropic, OpenAI, or Gemini credentials.
Strict Zero-Retention Privacy: Requests pass through in-memory streams with zero prompt caching or disk storage.
Drop-in Setup (Zero Code Changes)
Just swap the Base URL in your IDE or agent config:
Cursor / Windsurf: Change OpenAI Base URL to https://api.cntxtrim.com
Cline / Roo Code / Continue.dev: Point endpoint to https://api.cntxtrim.com
SDKs (Python / Node / LangChain): Pass base_url="https://api.cntxtrim.com"
Live Simulator & Free Tier (100k tokens): https://cntxtrim.com
Built with