HN Showcase
Back to browse
GitHub Repository

Context profiler for LLMs and AI agents - used to introspect context contents and reduce token costs

28 starsPython

Save your Claude/codex/copilot tokens by using context profiler

by iezhy·Jul 27, 2026·2 points·0 comments
Visit ProjectView on HN

AI Analysis

●●SolidSolve My ProblemShip It

CPU profiler for LLM contexts reveals exactly why your Copilot session is burning cash.

Strengths
  • •Proxy-based architecture requires zero code changes to existing agents like Claude Code or Copilot.
  • •Visualizes breakdown of system prompts, tool definitions, and file contents to identify bloat.
  • •Tracks context changes over time to help optimize recurring agent sessions for cost efficiency.
Weaknesses
  • •Requires installing CA certificates and running a local proxy, adding friction to the dev workflow.
  • •Primarily a diagnostic tool; users must manually act on insights to actually reduce costs.
Target Audience

Developers using AI coding assistants, AI infrastructure engineers

Similar To

LangSmith · Helicone · Braintrust

Post Description

Most devs using AI coding tools don't even imagine how much information they put in the context window - tools, MCPs, tool responses, active skills - all of which take up some real estate. Sometimes updating a tool description or trimming api response can give significant savings - and profiler helps to measure that

Similar Projects