Back to browse
GitHub Repository

On-premises proxy that catches runaway AI agent loops before they drain your API budget

1 starsGo

NakshGuard – on-prem proxy that stops AI agent loops

by PujanMirani·Jun 16, 2026·1 point·0 comments

AI Analysis

●●SolidSolve My ProblemShip It

Context velocity detection catches error-append loops before other budget guards do.

Strengths
  • Four detection layers including context velocity for the common error-append loop pattern.
  • Shadow mode lets you calibrate thresholds against real traffic before enforcement.
  • Hot config reload via SIGHUP means no downtime when tuning rules.
Weaknesses
  • Only supports OpenAI and Anthropic APIs; no support for other LLM providers yet.
  • Per-agent tracking requires X-Agent-ID header discipline across your codebase.
Target Audience

Teams running AI agents with API cost concerns

Similar To

LLM Guard · Portkey · LiteLLM

Similar Projects

AI/ML●●●Banger

LoopGain – Cut agent API spend by measuring when loops stop improving

1921 electrical engineering oscillator math stops AI loops at convergence.

Big BrainWizardry
fitz2882
216d ago
Developer Tools●●Solid

Agent Firewall – Go proxy to kill LLM death spirals

Wire-protocol circuit breaker for agents when LangSmith costs too much.

Solve My ProblemShip It
wuweiaxin
213mo ago
Security●●●Banger

ClawShield – Open-source security proxy for AI agents (Go, eBPF)

Defense-in-depth AI agent firewall: proxy + eBPF kernel + three-tier injection detection.

WizardrySolve My Problem
sleuthco
223mo ago