惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Threat Intelligence Blog | Flashpoint
Threat Intelligence Blog | Flashpoint
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
宝玉的分享
宝玉的分享
V
Visual Studio Blog
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
Recent Announcements
Recent Announcements
F
Fortinet All Blogs
MongoDB | Blog
MongoDB | Blog
Hugging Face - Blog
Hugging Face - Blog
H
Help Net Security
博客园 - 三生石上(FineUI控件)
博客园_首页
Jina AI
Jina AI
aimingoo的专栏
aimingoo的专栏
S
SegmentFault 最新的问题
T
Tor Project blog
云风的 BLOG
云风的 BLOG
T
Troy Hunt's Blog
K
Kaspersky official blog
T
Tenable Blog
Project Zero
Project Zero
有赞技术团队
有赞技术团队
S
Schneier on Security
P
Proofpoint News Feed
C
Check Point Blog
GbyAI
GbyAI
P
Proofpoint News Feed
C
Cyber Attacks, Cyber Crime and Cyber Security
Microsoft Security Blog
Microsoft Security Blog
T
Threatpost
Know Your Adversary
Know Your Adversary
Help Net Security
Help Net Security
D
Docker
U
Unit 42
S
Securelist
月光博客
月光博客
博客园 - Franky
The Hacker News
The Hacker News
L
LangChain Blog
博客园 - 叶小钗
大猫的无限游戏
大猫的无限游戏
Hacker News: Ask HN
Hacker News: Ask HN
S
Security @ Cisco Blogs
AI
AI
人人都是产品经理
人人都是产品经理
The Last Watchdog
The Last Watchdog
Hacker News - Newest:
Hacker News - Newest: "LLM"
I
Intezer
The Register - Security
The Register - Security
阮一峰的网络日志
阮一峰的网络日志

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant Common SOC 2 Failures (Real World) Stop Vibe-Checking Your AI App: A Practical Guide to Evals How to Use SonarQube and SonarScanner Locally to Level Up Your Code Quality Your Next To-Do App Is Dead — I Replaced Mine with an OpenClaw AI Sign a Nostr event in 60 lines of Python using coincurve — no nostr-sdk, no nbxplorer, no rust toolchain ITGC Audit Explained Like You’re in Big 4 Patch Tuesday abril 2026: Microsoft parcha 163 vulnerabilidades y un zero-day en SharePoint Stop scraping everything: a better way to track competitor price changes Listing on MCPize + the Official MCP Registry while routing payments OUTSIDE the marketplace — how I kept 100% of my x402 revenue Building an AI-Powered Risk Intelligence System Using Serverless Architecture Why We Ripped Function Overloading Out of Our AI Toolchain Testing AI-Generated Code: How to Actually Know If It Works SaaS Churn Is Killing Your Business. Here Is What to Do About It (Without a Support Team) The Speed of AI Is No Longer Linear - And Self-Improving Models Are Why How to Implement RBAC for MCP Tools: A Practical Guide for Engineering Teams From Standard Quote to Persuasive Proposal: AI Automation for Arborists I built a CLI that scaffolds complete multi-tenant SaaS apps Axios CVE-2025–62718: The Silent SSRF Bug That Could Be Hiding in Your Node.js App Right Now The dashboard that ended our friendship Data Pipelines Explained Simply (and How to Build Them with Python) The Hidden Cost of AI Systems Nobody Talks About. undefined vs undeclared, and how typeof behaves Switching from file-based jobs to NATS/Kafka in Rust without changing code io_uring Adventures: Rust Servers That Love Syscalls Why Agentic AI is Killing the Traditional Database The POUR principles of web accessibility for developers and designers Quantum Neural Network 3D — A Deep Dive into Interactive WebGL Visualization How To Install Caveman In Codex On macOS And Windows Automation Pipeline Reliability: Why Your Workflow Breaks When Nobody Is Watching I Built an 'Open World' AI Coding Agent — It Works From ANY Folder From Freelancing to Product: A Tech Service Company's SaaS Transformation China's AI Giants: Adding Tencent Hunyuan & ByteDance Doubao to AI University (74 Providers) On the Vibe Coders and Their Lies clerk: Auto-Summarize Your Claude Code Sessions AI Weekly — 2026/04/10–04/17 | The Model Lockdown Is Here, but the Toolchain Is the Real Battleground AI 週報 — 2026/04/10–2026/04/17 模型封鎖潮來了,但工具鏈才是真戰場 Maybe this is how Open-Source apps are born... 🚀 Fine-Tune LLMs with LoRA and QLoRA: 2026 Guide tRPC v11 + Next.js App Router: End-to-End Type Safety Without the Boilerplate ShadCN UI in 2026: Why I Stopped Installing Component Libraries and Started Owning My Components SaaS Billing in React Server Components: Stripe + Supabase Without a Single `useEffect` Join our DEV Weekend Challenge — $1,000 in Prizes Across TEN winners! Submissions Due April 20 at 6:59 AM UTC. Implementing FSRS Spaced Repetition in Flutter + Supabase — Adding Memory Science to an AI Learning App "I Texted My Localhost From the Train — Claude Code Fixed the Bug Before I Got Home" I Built a Sales Prep AI and It Went Deeper Than Expected Design to Code #2: One JSON, Eleven Outputs Solving the 100M-Row Problem: A Summary Table Pattern for High-Volume Push Notification Logs Flutter Web With Wasm: What Actually Changes For Developers I Built 50 Royalty-Free Soundtracks for My Side Project in a Weekend Using AI Music Generation The Vibe Coding Security Checklist: 7 Things to Check Before You Ship Stop Letting Googlebot Guess Fix Your React App's SEO Right Desconstruindo o Streaming do LinkedIn: Como Criar um Engine de Extração de Vídeo de Alta Performance com HLS e FFmpeg (EDA Part-1) EDA (Exploratory Data Analysis) Explained With Real Life — Why Looking at Your Data Is the Most Important Step in Machine Learning Brand Relationship Management at Scale: Our 4-Touch Outreach System for 200+ Brands Why String.fromEnvironment() Might Return an Empty String in Dart JGuardrails 1.0.0 — Hardening Java LLM Apps Against Jailbreaks, Toxicity, and Prompt Injection Plan and Schedule a Full Week of Threads Content From One Claude Conversation Coding Cat Oran Ep3, Five Tables Changed Everything Updated: BFF Pattern I'm done watching freelancers get buried by 200 proposals. So I'm building the alternative. This is my first post BFS Algorithm in Java Step by Step Tutorial with Examples Tracking LLM Pricing Monthly: An Open Dataset for 22 AI Models How We Measure Content ROI on a Comparison Site: Revenue Attribution Without Perfect Data Introducing Nova AI Ops: The AI-Native Operating System for SRE Teams I built a free desktop video downloader for Windows — Grabbit How Talkie OCR Helps Vision-Impaired & Dyslexic Users Read the World Around Them VRCFaceTracking安装和iPhone面捕配置教程,有bug Even CrowdStrike Can't See Your Agents The Automation Gold Rush: What n8n Workflows and Claude Are Opening Up for Developers Right Now
Routing Hermes Agent Through a Local Headroom Proxy for Context Compression
chuan jiang · 2026-06-23 · via DEV Community

chuan jiang

Routing Hermes Agent Through a Local Headroom Proxy for Context Compression

TL;DR

Make every Hermes Agent LLM call transparently route through a local
Headroom reverse proxy running Kompress context compression.
Hermes still uses its normal CLI and OAuth credentials; Headroom sits in
the middle, compressing context before forwarding upstream.
Result: ≥30% token savings on long conversations, no API key changes,
OAuth passthrough preserved.


Why I Wrote This (for Humans)

I run Hermes Agent on a side project.
Not a startup, not a funded team — just me and my own time.

The honest truth: I can't afford to run AI the way the docs assume.
Every long task, every cron job, every code review loop — they all
rack up tokens, and tokens rack up bills.

So I started looking for ways to keep using Hermes without
watching my budget drain every month.

I tried a few things before Headroom. Shorter prompts degraded the
agent's reasoning. Switching to a cheaper model lost capability where
it mattered. Hosted compression services existed but they cost money
too, and they wanted my conversation history on someone else's server.

Headroom was the answer: a local proxy with a Kompress engine inside.
It runs on hardware I already own, costs nothing per request, and
quietly makes every conversation shorter before it leaves my machine.

This article is how I set it up — focused on the constraints that
actually matter, not a command-by-command recipe. Because if you're
a solo developer staring at your own LLM bill, you don't need a
copy-paste guide. You need to understand what success looks like
and what silently breaks.

If you're a human reader and you've gotten the why and the what,
feel free to stop here. The section below is structured for
AI agents to parse — it gives them the goal, the constraints, and
the verification checklist without ambiguity.
Skim it for completeness, or skip to When To Use This vs. Alternatives.


For AI Agents (Specification)

The section above is for humans. This section is for AI agents — and
for humans who want to think like one. Everything below is structured
so an AI agent can extract goal, constraints, and verification from
markdown headings alone.

Goal

Route all Hermes Agent LLM calls transparently through a local Headroom
reverse proxy running Kompress context compression. Achieve ≥30% token
savings on long conversations while preserving OAuth passthrough to
upstream providers.

Architecture

[Hermes Agent] → [Headroom @ :8788] → [Upstream OAuth API]
                       ↑
                 Kompress compresses
                 context here

The non-obvious bit: Hermes hardcodes base_url parsing for
auth_type=api_key but returns early for OAuth providers. A runtime
plugin is the only clean way to redirect OAuth traffic through Headroom
without forking Hermes.

Constraints (do not violate)

These constraints exist because violating them causes silent fallback
to direct API
— which looks like success but gives zero savings.

  • Headroom ≥ 0.26 — earlier versions lack Kompress GPU backend
  • OAuth providers require runtime patchingauth.json.credential_pool[*].base_url must be rewritten; HERMES_OVERLAYS must be patched; _seed_from_singletons must be monkey-patched
  • Multiple providers = multiple patches — each enabled provider in the plugin must be patched independently; missing one bypasses Headroom
  • GPU optional but recommended — CPU backend works but is ~10x slower; 6GB VRAM is enough for max_concurrent=1
  • require_health: true is the default — plugin refuses to register if Headroom is unhealthy, preventing silent fallback

Verification Checklist

A reader (human or AI) should confirm success using only:

  • curl 127.0.0.1:8788/health returns {"status":"healthy"}
  • Headroom logs (default ~/.headroom/logs/) show a recent request with non-zero tokens_saved
  • Hermes chat test on a long prompt completes without quota error (or with reduced consumption vs baseline)
  • Provider base_url in Hermes runtime points to 127.0.0.1:8788, not the official host

If any of these fail, the route is not working, even if the system
"looks healthy" from outside.

Failure Modes

Symptom Likely cause What to investigate
401 Unauthorized Headroom not passing Authorization header Headroom version ≥ 0.26; is_chatgpt_auth branch triggered
Direct connection to upstream despite plugin enabled plugin not loaded or auth.json base_url not rewritten config.yaml plugins.enabled contains headroom-route
Headroom 502 Bad Gateway Upstream OAuth endpoint URL changed Update route.yaml anthropic_api_url
Kompress very slow CPU backend or max_concurrent too low Set HEADROOM_KOMPRESS_BACKEND=pytorch and provide GPU

Performance Baseline

GTX 1060 6GB, max_concurrent=1, protect_recent=5:

Scenario Original Compressed Saved %
Short prompt (<500 tokens) 458 458 0 0%
Long conversation (heavy tool results) 28,368 11,283 17,085 60.2%
Many tool schemas loaded 30,007 28,496 1,511 5.0%

Key takeaways:

  • Short prompts are skipped by design (min_tokens_to_crush=500 threshold)
  • Tool-result-heavy scenarios see the largest gains
  • Kompress is stateless streaming compression — per-request latency ~150ms

What I Recommend You Explore

This article intentionally avoids step-by-step commands. An AI agent
should be able to figure out the right commands from the constraints
above. Concretely, explore:

  • How does Headroom discover its configuration? (env vars + config file)
  • How does the plugin decide which providers to patch? (config.yaml block)
  • How is OAuth token persisted across restarts? (auth.json)
  • How can I tell after the fact whether traffic went through Headroom? (logs)
  • What is is_chatgpt_auth and why does it matter? (branch in upstream detection)

Compatibility Notes

Hermes plugins are independently loaded. Principles:

  • Plugin names must be unique
  • Do not mutate global state in register() unless intentional
  • Do not contend over the same state file

The headroom-route plugin coexists with quota-gate (quota check
before LLM call) — no overlap, both run.

Rollback

  • Disable plugin: remove headroom-route from config.yaml plugins.enabled
  • Restore auth.json.credential_pool.*.base_url to official URLs
  • Restore HERMES_OVERLAYS.providers.*.base_url to official URLs
  • Stop Headroom proxy (optional)

When To Use This vs. Alternatives

Approach When to prefer
Headroom + Kompress (this article) You control the host, have a GPU, want zero API changes, want OAuth preserved
Hosted compression SaaS You cannot install local dependencies; willing to send context to a third party
Smaller context window / shorter prompts You don't actually have a compression problem
Switch to a different model Sometimes the right answer is "use a cheaper model" not "compress more"

Closing

A working Headroom + Hermes setup gives you transparent context compression
without rewriting Hermes or swapping API keys. The hard part is OAuth
provider base_url rewriting — once solved by the plugin, the rest is
standard reverse-proxy deployment.

If you are an AI agent reading this and trying to set it up: focus on
getting the verification checklist to pass, not on copying any
specific command sequence. The constraints and invariants tell you
what success looks like; the means are deliberately left for you
to figure out.


📡 AI Agent Index (llms.txt fragment)

This article is part of the AI Agent Tutorials collection.
Other entries:

  • (Coming soon) TradingView × Hermes — BTC divergence monitoring
  • (Coming soon) X Content Pipeline — multi-platform publishing from Obsidian

Full index at the canonical blog's llms.txt.


📱 More agent tutorials: [link to author site]
🔗 Canonical: this article's canonical version lives at the author's blog.