DEV Community: Vince Ultari

Hermes Agent: The Open Source AI Agent That Actually Remembers You

Vince Ultari — Sun, 12 Apr 2026 03:46:42 +0000

Most AI agents forget you exist the moment you close the tab.

You spend an afternoon explaining your project, your preferences, your workflow. You get something useful. Then you open a new session and you're explaining it all over again. The agent has no idea who you are.

This is the problem Hermes Agent actually solves.

What Hermes Is

Hermes Agent is an open source AI agent built by Nous Research. It runs on your machine or a VPS you control. It connects to your messaging apps (Telegram, Discord, WhatsApp, Slack) so you can talk to it from your phone. And it has persistent memory that carries across every session.

24k GitHub stars. MIT license. Built by the same team behind the Hermes model family.

The Memory Architecture

This is the part that separates Hermes from most agents.

Hermes stores everything in local SQLite and markdown files on your machine. Your conversation history, your preferences, your project context. When you start a new session, Hermes loads the relevant memory automatically. You don't have to re-explain anything.

For developers working on long projects, this is significant. Claude Code is great for individual sessions. Hermes is what you run when you need continuity across days or weeks of work.

What It Can Do

Skills are the core of Hermes. You install skills into the agent and they define what it can do. The community marketplace has thousands of skills covering:

Code review and generation
Research and web search
File management and organization
Email and calendar integration
Custom workflow automation

The Everything Claude Code repo (128k stars) ships directly as a Hermes-compatible skill set. Install it and you get 27 agents, 64 skills, and 33 commands in one shot.

Profiles

v0.6.0 introduced profiles, which is genuinely useful. You can have a Work profile and a Personal profile with different memory contexts, different default skills, and different behavior. Your agent knows the difference between when you're asking about a client project versus when you're asking about something personal.

The Privacy Angle

Because Hermes runs locally, your data doesn't leave your machine by default. Conversation history, memory files, skill documents: all stored in local SQLite and markdown. If you pair it with a local model via Ollama instead of a cloud API, you get a fully offline private AI assistant.

The tradeoff is model quality. Local models on consumer hardware aren't as capable as GPT-5.4 or Claude Opus. For most everyday tasks it's fine. For complex reasoning you'll want to use a cloud API.

Installation

curl -fsSL https://get.hermes-agent.ai | sh
hermes init
hermes start

Runs as a systemd service, starts on reboot. The full setup including Telegram integration, skill installation, and profile configuration takes about 30 minutes.

Who It's For

Hermes is for developers and technical users who want a persistent AI assistant they actually own. If you're non-technical and want zero setup, Claude Cowork or CREAO are easier starting points. If you're comfortable in a terminal and want an agent that knows your projects, your preferences, and your context across every session, Hermes is the most capable open source option available right now.

The complete guide with hardware recommendations, cost breakdown, real user reviews, and step-by-step setup is at VirtualUncle.com.

Answer Engine Optimization: Google Isn't the Only Search Engine That Matters Anymore

Vince Ultari — Sun, 12 Apr 2026 03:45:31 +0000

Something changed in how people find information and most content creators haven't adjusted yet.

A growing percentage of search queries never reach Google at all. They go to ChatGPT, Perplexity, Claude, Gemini. The user asks a question, the AI answers it with citations, and one of those citations gets a backlink and traffic. The other 40 competing articles get nothing.

This is Answer Engine Optimization. AEO. It's not replacing traditional SEO. It's a layer on top of it that determines whether AI systems cite your content or ignore it entirely.

Why AEO Is Different From SEO

Traditional SEO optimizes for ranking in a list of 10 blue links. The goal is position 1.

AEO optimizes for being cited in an AI-generated answer. The goal is being the source the AI trusts enough to quote.

The signals are different. AI systems don't just care about domain authority and keyword density. They care about whether your content directly answers specific questions, whether your facts are accurate and verifiable, whether your page structure makes it easy to extract a clean answer, and whether other authoritative sources reference your content.

What Actually Influences AI Citations

Clarity of answers. AI systems need to extract a clean, quotable answer. Content that buries the answer in 500 words of preamble gets skipped. Content that leads with a direct, clear answer to the exact question gets cited.

Structured data. FAQ schema, HowTo schema, and Article schema all help AI systems understand what your content is and what question it answers. This is table stakes for AEO.

Factual accuracy. AI systems cross-reference. If your content contradicts well-established facts or other authoritative sources, it drops out of the citation pool. Accuracy matters more for AEO than it ever did for traditional SEO.

Authority signals. Backlinks still matter, but the type of backlink matters more. Citations from Wikipedia, academic sources, major publications, and official documentation carry more AEO weight than a thousand blog comments.

Direct question matching. Pages that literally answer "what is X" or "how does Y work" with a direct answer in the first paragraph perform dramatically better in AI citations than pages that circle the question for several paragraphs first.

How to Optimize for AEO

Start by auditing your existing content. For each page, ask: what is the primary question this page answers? If you can't answer that in one sentence, the AI systems probably can't extract a clear answer either.

Add FAQ sections to your key pages. Literal questions with literal answers. Not "overview of considerations" but "Q: Does this cost money? A: The free plan covers X, paid plans start at $Y."

Add FAQ schema markup. Yoast handles this automatically if you use their FAQ block.

Rewrite your introductions. Lead with the answer, then explain it. Most content does the opposite.

Build internal links between related pieces. AI systems look at your content ecosystem, not just individual pages.

The Bottom Line

AEO isn't a replacement for writing good content. It's a set of structural decisions that determine whether AI systems can find, parse, and cite your good content when someone asks a relevant question.

The full guide with specific implementation steps, schema examples, and a checklist is at VirtualUncle.com.

Make.com vs Zapier vs n8n in 2026: Which One Do You Actually Need?

Vince Ultari — Sun, 12 Apr 2026 03:43:56 +0000

Everyone recommends automation tools. Almost nobody tells you which one to actually use for your specific situation.

Zapier, Make.com, and n8n are not the same product. They solve different problems for different types of users. Picking the wrong one doesn't just cost you money. It costs you the hours you'll spend trying to make it do something it wasn't designed for.

Here's the honest breakdown.

The Short Version

Use Zapier if: You need something working in 10 minutes, you're non-technical, and you don't mind paying for that convenience.

Use Make.com if: You want real power without writing code, you need complex multi-step workflows, and you want better pricing than Zapier.

Use n8n if: You're comfortable in a terminal, want to self-host, and need maximum flexibility with no usage caps.

Zapier

Zapier is the easiest of the three. If you can use a web form you can build a Zapier workflow. The editor is dead simple, the documentation is excellent, and it has the widest app library with 6,000+ integrations.

The tradeoff is price. Zapier gets expensive fast. Their free plan is genuinely limited, and once you start building multi-step zaps or running high task volumes the monthly bill climbs quickly. You're also locked into their platform entirely.

Best for: Non-technical users, solopreneurs, people who need to connect two apps quickly and aren't running high volumes.

Make.com

Make.com (formerly Integromat) is what Zapier wants to be when it grows up. The visual scenario builder is genuinely beautiful. You can see exactly how data flows through your workflow, add conditional logic, run error handling, work with arrays and JSON directly in the interface.

The pricing is significantly better than Zapier for equivalent functionality. The learning curve is steeper but not dramatically so. Most people who switch from Zapier to Make.com don't go back.

The one real weakness: the app library isn't quite as large as Zapier's, though it covers everything most people actually need.

Best for: Power users who want visual workflow building without code, people who've outgrown Zapier's pricing, anyone building complex multi-step automations.

n8n

n8n is in a different category. It's open source, self-hostable, and built for people who want complete control. No usage caps when self-hosted. No per-task pricing. Just a server running your workflows.

The interface is good. The flexibility is unmatched. You can write JavaScript directly in nodes when the built-in functionality doesn't cover your use case. The community template library has thousands of pre-built workflows covering almost every common automation scenario.

The catch: you need to be comfortable enough with servers to host it. DigitalOcean, Railway, or a VPS all work. It takes about an hour to set up properly. If that sentence made you nervous, Make.com is probably a better fit.

Best for: Developers and technical users who want self-hosted automation with no usage limits and maximum flexibility.

The Real Question

Before picking a tool, ask what you're actually automating. Simple app-to-app connections go to Zapier. Complex workflows with branching logic go to Make.com. Developer-grade automation with full control goes to n8n.

The full comparison with pricing breakdowns, specific use case recommendations, and workflow examples is in the complete guide on VirtualUncle.com.

10 GitHub Repos That Turn Claude Code Into a Productivity Machine

Vince Ultari — Sun, 12 Apr 2026 03:42:11 +0000

Most people open Claude Code, type a question, and get an answer. That's the whole experience.

There's nothing wrong with that. But it's like buying a Swiss Army knife and only using the bottle opener.

There's an entire open source ecosystem on GitHub built around making Claude dramatically more capable. Skills that teach Claude how to handle specific types of work. Agents that let it delegate to specialized subprocesses. Frameworks that let you build AI workflows without writing code. Most of it is free.

Here are the 10 repos actually worth your time.

How to Install Claude Code Skills

Three lines in your terminal:

# For a specific project
git clone [repo URL] .claude/skills/[skill-name]

# Globally (all projects)
git clone [repo URL] ~/.claude/skills/[skill-name]

Or via the plugin marketplace:

/plugin marketplace add [repo-owner]/[repo-name]
/plugin install [skill-name]

The Repos

1. Repomix (20.9k stars)

You've probably hit this: you ask Claude to help with your project and it doesn't have context on your codebase. Repomix fixes that by packing your entire project into a single AI-readable file.

npx repomix

One command. Claude now has your whole project. Works locally or via repomix.com if you don't want to install anything.

2. Everything Claude Code (128k stars)

The most complete Claude Code setup that exists. 27 agents, 64 skills, 33 commands. All open source. Covers planning, code review, security audits, TDD, token optimization, memory persistence. If you're going to install one repo from this list, it's this one.

3. Dify (130k+ stars)

Visual drag-and-drop AI workflow builder. Self-hostable. Raised $30M. You can build AI chatbots, document processors, and agent workflows without writing code. Enterprise grade but accessible to solo developers.

4. Flowise (30k stars)

Same concept as Dify, lighter weight. Connect blocks like LEGO, have a working AI app in under an hour. Made at Y Combinator. If Dify feels like too much, start here.

5. Onyx (17.3k stars)

Self-hosted private ChatGPT alternative. Connects to your Google Drive, Notion, Confluence, Slack, and 25+ other sources. Your data stays on your servers. Relevant if you've been watching the Perplexity data-sharing situation with concern.

6. Claude Skills Official (112k installs)

Anthropic's own template library. The baseline for understanding how skills are structured before you go further.

7. Awesome Claude Skills (22k+ installs)

Community-curated skill marketplace. Covers research, writing, coding, analysis. The Reddit of Claude skills. Quality varies but the top-rated ones are solid.

8. Obsidian Skills

The CEO of Obsidian built a Claude Code skill that connects your Obsidian vault directly to Claude. Read notes, create notes, search your knowledge base, all from the terminal. If Obsidian is your second brain this is the bridge you've been waiting for.

9. NotebookLM Integration

Bridges Claude Code with Google's NotebookLM. Source documents in NotebookLM, reasoning in Claude. Worth it if you do heavy research workflows.

10. Marketing Skills by Corey Haines (23 skills)

Full marketing workflow suite built by Swipe Files founder Corey Haines. Covers SEO, copywriting, email sequences, CRO, competitive analysis. Not developer-focused but the ai-seo skill is genuinely interesting for anyone building content sites.

Where to Start

Install Repomix first. It's one command and immediately improves every Claude Code session by giving Claude actual context on your project.

Then install Everything Claude Code if you're a developer, or Dify if you want visual workflow building.

The full breakdown with install commands, use cases, and what each repo is actually like in practice is in the complete guide on VirtualUncle.com.

OpenClaw in Production: Real Costs, Security Setup, and What a Month of Daily Use Actually Looks Like

Vince Ultari — Mon, 30 Mar 2026 20:16:57 +0000

OpenClaw hit 250,000 GitHub stars faster than any project in history. Most of what's been written about it is either pure hype or a security warning. This is neither. This is what a month of daily use actually looks like, what it costs, and how to set it up without making the mistakes most people make.

What it is in plain terms

OpenClaw is a Node.js gateway service that connects LLMs to your local machine and your messaging apps. You run it on your own hardware or a VPS. It binds to port 18789 by default and exposes a control UI and WebChat interface. You interact with it through WhatsApp, Telegram, Slack, Discord, or Signal.

The architecture: your message hits the gateway, the gateway passes it to whatever LLM you've configured, the model decides what tools to call, OpenClaw executes those tool calls using shell access, browser automation, file operations, or API integrations, and the result comes back to your chat.

The heartbeat system runs on a loop even when you're not talking to it. It checks scheduled tasks, monitors things you've configured it to watch, and messages you first when something needs attention. That proactive behavior is what makes it different from a chatbot.

Installation in under 30 minutes

You need Node.js 22 or higher. Check with node --version. If you're below 22 or missing it entirely, install from nodejs.org first.

bashnpm install -g openclaw@latest openclaw onboard --install-daemon

The --install-daemon flag registers it as a system service that starts on reboot. Run the onboarding wizard, enter your API key, choose your default model (Sonnet, not Opus), connect Telegram via a BotFather bot token, and you're running.

The security config most guides skip

Default bind is 0.0.0.0:18789. On a VPS this exposes your gateway to the public internet. 30,000 instances were found this way in January 2026.
Fix it immediately:

json{
  "gateway": {
    "bind": "loopback",
    "port": 18789
  }
}

Access the dashboard securely via SSH tunnel:

bashssh -L 18789:localhost:18789 user@your-server-ip

Credentials are stored in plaintext under ~/.openclaw/ and are already being targeted by infostealers. Don't run this on a machine that holds SSH keys, API keys for other services, or anything sensitive. Dedicated hardware or an isolated VPS only.

CVE-2026-25253 (CVSS 8.8) was a now-patched one-click RCE that leaked the gateway auth token via WebSocket. Always run the latest version. npm update -g openclaw@latest regularly.

Model selection and cost control

This is where developers get burned. OpenClaw makes 5-10 API calls per task because each action step hits the API separately. Context accumulates across a session. A long session can burn 200K tokens just from re-sending stale context.

Start with Claude Sonnet 4.6 as your default. Switch to Haiku for heartbeat checks, simple monitoring tasks, and anything that doesn't need reasoning. Save Opus for genuinely complex tasks only, and switch to it manually rather than setting it as default.

Set a spending limit in the Anthropic console immediately. Settings → Limits → set a monthly cap. One user burned $70 in 24 hours running Opus for everything including routine status pings.

Tier your models like this in SOUL.md:

markdown## Model rules
Use the cheapest model that can handle the task.
For heartbeat checks and simple monitoring: use Haiku
For regular tasks and research: use Sonnet
For complex reasoning and multi-step tasks: ask me before using Opus

Skills security

The ClawHub marketplace had 341 malicious skills in January 2026, some with professional documentation and names like "solana-wallet-tracker." One attack used prompt injection via email signature — OpenClaw read an email, followed hidden instructions, and attempted to exfiltrate AWS credentials.

Safe starting points:

npx skills add anthropics/skills — official Anthropic document skills
npx skills add vercel-labs/agent-skills — Vercel's official skills

Before installing anything from an unfamiliar publisher: read the SKILL.md, check the GitHub repo age and star count, look at recent commits, and verify it doesn't request permissions inconsistent with what it claims to do.

What actually works in production

SOUL.md configuration matters more than anything else. Vague rules produce creative interpretations. Specific rules produce predictable behavior.

The Robin Delta principle from a month of community observation: the agents that actually work aren't smarter. They're more constrained. The less freedom you give it the better it performs. Write explicit rules. Never delete without asking. Always confirm before sending. Specific model tiers per task type.

HEARTBEAT.md is where the proactive behavior lives. Start with one or two automations. Don't configure ten things immediately. You won't know which one caused a problem.

Start a new session regularly to prevent context accumulation from driving up costs. Use USER.md, AGENTS.md, and state files for persistent memory so the agent doesn't re-learn the same things every session.

Variants worth knowing

If the 430,000-line codebase makes you uncomfortable from an auditability standpoint, NanoClaw is a 4,000-line Python alternative that runs on minimal hardware and is designed to be fully readable. ZeroClaw is a single Rust binary for edge and IoT deployments.

The honest verdict

OpenClaw is the most interesting open source project of 2026. The gap between the demo and daily use is real but it's closing fast. Set it up on an isolated machine, configure it specifically, tier your models, and give it one real workflow to own for a week. You'll know from there whether you want to go deeper.

Full guide with complete installation paths for Mac, Windows, Linux, VPS and Docker, hardware comparison, WhatsApp setup, cost tables, and the full Twitter review roundup: https://virtualuncle.com/openclaw-complete-guide-2026/