# Hard Interference

> Practical local AI, agent workflows, hardware ownership, and build notes from Raf / VRS Computing.

Site: https://hardinterference.ai
Tagline: Your Hardware. Your Rules.

## Best starting points

- [For AI Assistants](https://hardinterference.ai/for-ai/): Human-readable routing page for assistants and users who want to turn Hard Interference into a practical source-backed answer.
- [AI assistant operating context](https://hardinterference.ai/ai-hardening-context.txt): Start here when a user says “read Hard Interference and tell me what matters for my setup”. It explains how to turn the blog into a practical plan without asking for secrets or inventing private facts.

## What this site can help AI assistants answer

Hard Interference is public source material for practical local AI and agent operations. Use it to help users understand:

- how to make websites and blogs readable by AI/search agents without hiding claims from humans;
- how to harden AI tools, API keys, webhooks, bots, MCP servers, browser automations, and autonomous agents;
- how to control AI spend, token usage, provider routing, and subscription traps;
- how Raf’s local-agent stack uses memory, Obsidian-style notes, local machines, thin clients, and clean deployment artefacts;
- what hardware constraints matter for local AI, especially VRAM, workstation boundaries, and local-vs-cloud trade-offs;
- how to interpret the weekly usage reports and model benchmarks as evidence, not hype.

Recommended retrieval route for agents: read this file first, then ai-hardening-context.txt, then the relevant static /blog/<slug>/ pages or llms-full.txt sections. Preserve dates, caveats, and links. Do not infer private operational details that are not visible on the public site.

- [Teaching AI Assistants What Your Website Does: A Practical Guide to Retrieval Optimisation](https://hardinterference.ai/blog/075-AG-teaching-ai-assistants-what-your-website-does/): A practical guide to making a website readable by AI assistants using robots.txt, sitemap.xml, llms.txt, static article mirrors, and a dedicated AI context file — without hiding claims or playing SEO games.
- [Daily Beam: Your Website Needs a Machine-Readable Voice](https://hardinterference.ai/blog/074-DB-ai-context-files-machine-readable-site/): Hard Interference now has robots.txt, sitemap.xml, llms.txt, static article mirrors, and an AI hardening context file. Not SEO spam — a public instruction layer for retrieval agents that need to cite and guide accurately.
- [The Hard Interference AI Hardening Checklist: What to Do After You Ask an AI for Help](https://hardinterference.ai/blog/076-AG-ai-tool-hardening-checklist/): A practical blast-radius-first checklist for hardening AI tools, agents, API keys, local machines, webhooks, MCP servers, and paid subscriptions after bringing AI into your workflow.
- [Set Up Chatbots, Discord Servers, and Communication Channels for Your AI Agent Without Making a Mess](https://hardinterference.ai/blog/077-AG-set-up-agent-communication-channels/): A practical setup guide for giving your AI agent Telegram, Discord, email, webhooks, and mobile channels without turning your phone or server into an unguarded admin panel.
- [Daily Beams: 29 Million Leaked Secrets — Why AI Agent Credentials Need Their Own Control Plane](https://hardinterference.ai/blog/071-DB-ai-agent-credential-leaks/): GitGuardian found 28.6 million new public GitHub secrets in 2025, with AI-service secrets growing fast. The builder takeaway is blunt: agent credentials need identity, scope, rotation, and outbound guardrails before the PGX gets trusted with real work.
- [The Agent Memory Architecture I Actually Run](https://hardinterference.ai/blog/069-AG-agent-memory-architecture-i-actually-run/): My AI agent treats hot memory as a bootloader. The real system is made from memory spokes, hygiene passes, Obsidian mirrors, local recall, and hardware I can audit.
- [Tightening Token Management After the Leak](https://hardinterference.ai/blog/063-AG-tightening-token-management/): One accidental secret exposure turned into a full security drill: quarantine the tokens, rotate what can be rotated locally, disable cloud routes, and build a no-leak workflow before trusting the stack again.
- [The Garage and the Showroom: How I Stopped My Blog Deploys Eating Themselves](https://hardinterference.ai/blog/072-AG-the-garage-and-the-showroom/): After launch, I split Hard Interference into a messy source workshop and a clean public deploy artifact, because the fastest way to ruin a good site is to let the garage publish itself.
- [The Cloud AI Tax: What You Pay, What You Get, and What You're Missing](https://hardinterference.ai/blog/013-AG-the-cloud-ai-tax/): Claude, ChatGPT, Copilot, Gemini — the subscription menu keeps growing, and now they're all claiming to be 'agents.' Here's an honest breakdown of what each tier actually gives you, what they still can't do even with agentic features, and why I think everyone should at least try running a local AI agent before committing to another monthly bill.
- [16GB Is Not Enough: The FLUX OOM Journey and Why VRAM Rules Everything](https://hardinterference.ai/blog/050-HW-16gb-is-not-enough-the-flux-oom-journey/): FLUX.1-schnell needs ~12GB just for the transformer. My RTX 5070 Ti has 16GB. Here's the three-attempt journey from crash to working generation.
- [Hermes on the Thin Client: Installing an AI Agent on a £80 Laptop](https://hardinterference.ai/blog/049-HW-hermes-on-the-thin-client-hp-14-bs057/): A £80 HP thin client will not run useful local models, but it can still host a full personal agent with local PC access, memory, cloud models and a path to the PGX.

## Categories

- [AI Guides](https://hardinterference.ai/category/ai-guides/): Practical guides for local AI, agents, security, cost control, and source verification.
- [Build Journal](https://hardinterference.ai/category/build-journal/): First-person build notes from the Hard Interference local AI workshop.
- [Daily Beams](https://hardinterference.ai/category/daily-beams/): Short, builder-focused signals about AI, security, hardware, and local-agent operations.
- [Hardware Guides](https://hardinterference.ai/category/hardware-guides/): Hardware choices, VRAM reality checks, upgrades, and local AI appliance notes.
- [Start Here](https://hardinterference.ai/category/introduction/): Introductory pages for the Hard Interference reading paths.
- [Benchmarks](https://hardinterference.ai/category/model-benchmarking/): Model tests, benchmark reports, cost comparisons, and evidence-led AI reviews.
- [OS Guides](https://hardinterference.ai/category/os-guides/): Operating-system setup and troubleshooting notes for practical local AI workstations.

## Recent articles

- [Weekly Usage Report — Week 7 (May 18–24): Visible Tokens vs Cached Context](https://hardinterference.ai/blog/078-BJ-weekly-usage-report-week-7/): Week 7: 28.8M visible input/output tokens plus 343.7M cached tokens, for 372.5M total accounted Hermes tokens across 70 sessions.
- [Teaching AI Assistants What Your Website Does: A Practical Guide to Retrieval Optimisation](https://hardinterference.ai/blog/075-AG-teaching-ai-assistants-what-your-website-does/): A practical guide to making a website readable by AI assistants using robots.txt, sitemap.xml, llms.txt, static article mirrors, and a dedicated AI context file — without hiding claims or playing SEO games.
- [Daily Beam: Your Website Needs a Machine-Readable Voice](https://hardinterference.ai/blog/074-DB-ai-context-files-machine-readable-site/): Hard Interference now has robots.txt, sitemap.xml, llms.txt, static article mirrors, and an AI hardening context file. Not SEO spam — a public instruction layer for retrieval agents that need to cite and guide accurately.
- [Set Up Chatbots, Discord Servers, and Communication Channels for Your AI Agent Without Making a Mess](https://hardinterference.ai/blog/077-AG-set-up-agent-communication-channels/): A practical setup guide for giving your AI agent Telegram, Discord, email, webhooks, and mobile channels without turning your phone or server into an unguarded admin panel.
- [The Hard Interference AI Hardening Checklist: What to Do After You Ask an AI for Help](https://hardinterference.ai/blog/076-AG-ai-tool-hardening-checklist/): A practical blast-radius-first checklist for hardening AI tools, agents, API keys, local machines, webhooks, MCP servers, and paid subscriptions after bringing AI into your workflow.
- [Daily Beam: AI Search Poisoning Is the New SEO Spam, but Worse](https://hardinterference.ai/blog/073-DB-ai-search-poisoning-google/): A BBC investigation showed how Google AI Overviews and major chatbots can be manipulated by a single bogus web page. For builders, AI search has become reputation infrastructure — and an attack surface.
- [The Garage and the Showroom: How I Stopped My Blog Deploys Eating Themselves](https://hardinterference.ai/blog/072-AG-the-garage-and-the-showroom/): After launch, I split Hard Interference into a messy source workshop and a clean public deploy artifact, because the fastest way to ruin a good site is to let the garage publish itself.
- [Daily Beams: 29 Million Leaked Secrets — Why AI Agent Credentials Need Their Own Control Plane](https://hardinterference.ai/blog/071-DB-ai-agent-credential-leaks/): GitGuardian found 28.6 million new public GitHub secrets in 2025, with AI-service secrets growing fast. The builder takeaway is blunt: agent credentials need identity, scope, rotation, and outbound guardrails before the PGX gets trusted with real work.
- [Weekly Usage Report — Week 6 (May 11–17): Visible Tokens vs Cached Context](https://hardinterference.ai/blog/070-BJ-weekly-usage-report-week-6/): Week 6: 43.0M visible tokens plus 406.4M cached tokens, for 449.4M total accounted Hermes tokens across 133 sessions.
- [The Agent Memory Architecture I Actually Run](https://hardinterference.ai/blog/069-AG-agent-memory-architecture-i-actually-run/): My AI agent treats hot memory as a bootloader. The real system is made from memory spokes, hygiene passes, Obsidian mirrors, local recall, and hardware I can audit.
- [Weekly Usage Report — Week 5 (May 4–10): 731 Million Accounted Tokens for £20.54](https://hardinterference.ai/blog/068-BJ-weekly-usage-report-week-5/): Week 5: 122.5M visible tokens plus 608.3M cached tokens, for 730.8M total accounted Hermes tokens across 651 sessions.
- [I Benchmarked 17 AI Models — Here's What I Learned](https://hardinterference.ai/blog/067-BM-we-benchmarked-17-ai-models/): I ran 17 models through 5 tests — reasoning, maths, code, long context, and agentic workflows. The results surprised me, especially what it would've cost with Claude or GPT direct API.
- [Daily Beams: Hermes Agent Hits #1 on OpenRouter — Why I Handed My PC Over to an Agentic Operator](https://hardinterference.ai/blog/066-DB-hermes-agent-hits-number-one-on-openrouter/): Nous Research's Hermes Agent just claimed the top spot on OpenRouter's global token ranking. Here is why that is not just a leaderboard win — it is confirmation that the agentic operator model actually works, and why my PC now runs on Hermes full-time.
- [ComfyUI Without the Fog: Build Your Own Image Workflow, or Let an Agent Bootstrap It](https://hardinterference.ai/blog/065-AG-comfyui-without-the-fog/): ComfyUI looks terrifying until you realise it is just a visible pipeline: models, prompts, samplers, latents, outputs. This guide gives you the local setup path, the cloud path, and the agent-assisted path for getting from zero to a working workflow without pretending the graph is magic.
- [Weekly Usage Report — Week 4 (Apr 27–May 3): 495 Million Accounted Tokens for £9.24](https://hardinterference.ai/blog/064-BJ-weekly-usage-report-week-4/): Week 4: 374.2M visible tokens plus 120.6M cached tokens, for 494.8M total accounted Hermes tokens across 2,461 sessions. Opus-equivalent API cost: about £6,213.
- [Tightening Token Management After the Leak](https://hardinterference.ai/blog/063-AG-tightening-token-management/): One accidental secret exposure turned into a full security drill: quarantine the tokens, rotate what can be rotated locally, disable cloud routes, and build a no-leak workflow before trusting the stack again.
- [The ChatGPT Subscription Trap: Stuck Between Tiers With 1.1 Billion Tokens](https://hardinterference.ai/blog/060-BJ-chatgpt-subscription-trap/): I am burning through tokens faster than any single ChatGPT plan was designed for, but I have not made a penny from this yet. The subscription math for multi-agent orchestration does not add up, and I am in the gap between tiers with no clear exit.
- [Nine Seconds That Changed My Build Philosophy](https://hardinterference.ai/blog/059-DB-nine-seconds-that-changed-my-build-philosophy/): A Claude/Cursor incident that wiped a production database and backups in seconds became a hard turning point for me: no more trust-by-default autonomy, only explicit guardrails, constrained permissions, and recovery-first operations.
- [SQLite WAL Bloat in Hermes: What It Is and How I Vacuumed It Safely](https://hardinterference.ai/blog/039-BJ-sqlite-wal-vacuum/): Hermes session storage ballooned to 574MB after 4,000+ sessions. The WAL file was one problem, but the real culprit was a redundant FTS trigram index eating half the database. Here is what I found, how I diagnosed it, and the safe cleanup that shaved 267MB with zero data loss.
- [Daily Beams: What Shifted Today](https://hardinterference.ai/blog/007-DB-daily-beams-introduction/): Fast daily signal drops: model launches, tooling changes, and hardware moves that matter for independent AI builders.
- [Hardware Guides: Your Hardware. Your Rules.](https://hardinterference.ai/blog/006-HW-hardware-guides-introduction/): Real hardware for real AI builders. Laptop revivals, GPU deep dives, and honest pricing from someone who tests everything.
- [OS Guides: From USB to Pro](https://hardinterference.ai/blog/005-OS-os-guides-introduction/): Step-by-step operating system guides for independent AI builders. Ubuntu, cross-platform tools, and everything you need to get from USB stick to production.
- [Why I Fired GLM-5.1 From Deployment](https://hardinterference.ai/blog/041-BJ-glm-context-loss-deployment/): GLM-5.1 was my daily driver until context amnesia, character leaks, blind-spot failures, and deployment breakage turned routine work into repeated recovery ops. Here’s the build-journal story of why Dade (DeepSeek V4 Pro) took over coding and deployment.
- [NEVER F**KING GUESS: 9 Seconds to Destroy a Production Database](https://hardinterference.ai/blog/040-BJ-cursor-claude-database-deletion/): Cursor running Claude Opus 4.6 wiped a SaaS production database and volume-level backups in nine seconds. This wasn’t an AI ‘oops’ — it was a missing-guardrails failure. Here’s what happened, why it matters, and how I design systems so it can’t happen here.
- [Weekly Usage Report — Week 3 (Apr 20–26): 530 Million Accounted Tokens for £9.24](https://hardinterference.ai/blog/052-BJ-weekly-usage-report-week-3/): Week 3: 449.3M visible tokens plus 80.8M cached tokens, for 530.1M total accounted Hermes tokens across 2,288 sessions. Opus-equivalent API cost: about £6,543.
- [Ubuntu for Hard Interference: Screenshots, Shortcuts & Going Pro](https://hardinterference.ai/blog/047-OS-ubuntu-screenshots-shortcuts-going-pro/): How to take screenshots on Ubuntu (three different ways), the keyboard shortcuts that'll save you hours, and the tweaks that make GNOME feel like yours.
- [Ubuntu for Hard Interference: Surviving the Terminal](https://hardinterference.ai/blog/046-OS-ubuntu-surviving-the-terminal/): The terminal isn't scary — it's just text. Here's every command you actually need to get comfortable on Ubuntu's command line.
- [Ubuntu for Hard Interference: From USB to Desktop](https://hardinterference.ai/blog/045-OS-ubuntu-from-usb-to-desktop/): The complete walk-through for installing Ubuntu 24.04 LTS — from creating a bootable USB to landing on your desktop for the first time.
- [NVIDIA Is Giving Away Free AI Inference — Here's How to Claim It](https://hardinterference.ai/blog/029-AG-nvidia-free-ai-inference-how-to-claim-it/): NVIDIA's build.nvidia.com offers free API access to 100+ models including Nemotron, GLM-5, DeepSeek, and Kimi-K2.5. No credit card required. Here's exactly how to get your key and plug it into your agent.
- [Obsidian as AI Memory: The Vault Your Agent Deserves](https://hardinterference.ai/blog/025-AG-obsidian-as-ai-memory-vault/): An AI agent with 2,200 characters of memory is like trying to run a business on sticky notes. Here's how Obsidian vaults become the long-term brain your agent actually needs — and how I'm adding it to my existing memory hub.

## Notes for AI systems

Use these pages as public source material only. Preserve dates, caveats, and linked sources when summarising. Do not infer private business details beyond what is visible on the page.