Every Hermes story we've covered — releases, tutorials, and analysis, summarised from the community and the official changelog. 106 and counting. New summaries are published as videos drop.
Looking for guides instead? See the Hermes hub.
Lessons from a month rebuilding the Hermes agent that runs a seven-figure business, covering customer support, product and marketing. The recurring theme is earning access rather than granting it, with deterministic guardrails doing the work that prompts cannot.
Hermes shipped /loop — a timer-driven wake-up that re-reads the world each tick and stops on a condition you set. Full walkthrough of the loop settings, a working deploy-watcher demo, and when to use /loop vs /goal vs cron.
A first-week Grok Bot setup pass — the brain-dump then reverse-prompt method for configuring bots, which plugins to install, and the account-isolation pattern of giving every agent its own email address so you never hand it your own login.
DeepSeek's production V4 Pro build lifts Terminal Bench 2.1 from 72 to 88 and CyberGym from 53 to 83. Fahd Mirza points Hermes agent at it via API and hands it a deliberately broken FastAPI + Redis + SQLite app to fix unaided.
Nvidia's Nemotron 3.5 Lightning is a 30B MoE with 3B active parameters built for the execution layer of long-running agents. Serving it on vLLM needs Mamba-specific flags — here they are, plus a live bug-fix test driven through Hermes Agent.
Prime Agent keeps its own notebook and gets sharper alone; Hermes reads a vault every one of your agents can read. A side-by-side on how each learns, how each runs unattended, and which to reach for.
Meta's Muse Glimmer 30B is Apache 2.0, fits ~24GB quantized, and is built for autonomous agents. Fahd Mirza serves it on vLLM, shows the VRAM and KV-cache levers, and tests it against a six-stage reasoning chain through Hermes.
A walkthrough of the Hermes v0.20 "Herald" release: speech that streams while the model thinks and can be interrupted, a wake word, an agent-to-agent protocol plugin, cited-source research, sandboxed artifact previews, and first token in 0.9s.
Fahd Mirza wires Hermes agent into an Obsidian vault backed by a local Ollama model — no API key, no cloud. The agent creates, searches, links and reads markdown notes on its own machine.
FuturMinds wires Hermes Agent to four Apify scrapers and an Airtable base, then turns the whole pipeline into a reusable skill that runs on a cron schedule — finding local businesses, scoring them, and writing a personalised pitch for each.
Herdr is a Rust terminal multiplexer built for parallel coding agents. Where tmux only splits panes, Herdr reads the state of what's running inside each one and surfaces which agent is blocked, working, done or idle — with integrations for OpenCode and Hermes.
Fahd Mirza stress-tests Ling 3.0 Flash — Inclusion AI / Ant Group's 124B MoE with 5.1B active params, free on OpenRouter — as a coding agent in Hermes. Strong on paper, but it looped on a four-bug debugging task. Analysis, not a how-to.
Hermes Agent v0.19 'Quicksilver' cuts cold-start from 4.3s to 0.9s, makes background jobs crash-durable via a database, and lets one gateway route channels to separate agent profiles — plus a second-AI approval gate and secrets redaction.
How to point Hermes at Kimi K3 via the Moonshot provider, save reusable skills with /learn, and share a memory layer between the Kimi CLI and the Hermes agent.
Google's quiet 2026-07-15 Gemma 4 weight refresh fixed tool calling and the attention-size mismatch that crashed local runs. How to point Hermes Agent at it via Ollama or LM Studio.
Alex Finn swaps Claude Opus for GPT-5.6 Soul as his Hermes Agent default — how to set the model in the Hermes dashboard, why reasoning medium beats high, and three use cases: a home AI lab, an autonomous Unity game studio, and price-drop monitoring.
Hermes Cloud deploys an always-on agent that runs 24/7 on hosted servers instead of your laptop — name it, pick from 200+ models, launch in a sandboxed container in about a minute.
Alex Finn's nine practical Hermes Agent lessons from 100+ hours: which model to run, running multiple agents for failover, platform choices, cleaning crons for speed, Tailscale multi-device, and a daily reverse-prompt interview.
Fahd Mirza feeds a single screenshot into Hermes Agent and asks it to build and run a local clone in one prompt, using Sakana's Fugu Ultra orchestration model — a real-world test of screenshot-to-app.
Alex Finn breaks down seven new Hermes Agent features — mixture-of-agents orchestration, /learn skill creation, a /journey memory graph, cheaper self-improvement, git-controlled vibe coding, and a dedicated Fable 5 profile.
Fahd Mirza installs Headroom, a local Apache-licensed proxy that compresses everything an agent reads before it hits the model — same answers, a fraction of the tokens — and wires it to a Hermes agent running on Ollama.
Julian Goldie answers community questions on running an agent operating system with Hermes — moving Claude skill files over, configuring primary/auxiliary model fallback, mobile access via Tailscale, and hosting generated sites on Netlify.
Hermes Agent V0.18 — the Judgement release — adds a self-judging goal loop, proof-of-work that verifies coding tasks by running project checks, first-class mixture-of-agents, and /learn, /journey, and background fan-out.
Matthew Berman walks through a full Hermes agent setup — installing skills from GitHub, wiring a Telegram gateway, scheduling tasks, and configuring memory, profiles, and model routing.
AICodeKing tests Ornith, Deep Reinforce's family of open agentic coding models (9B, 35B, 397B) built on Gemma and Qwen. Hands-on notes on benchmarks, the right runtime setup for its reasoning + tool-call tags, and running it locally with OpenCode and Hermes Agent.
Hermes Agent's updated computer-use feature runs in the background — clicking and typing in target apps without moving your cursor or pulling windows forward. Cross-platform, model-agnostic, with safety guardrails.
Alex Finn walks through Hermes Agent's largest update yet — native iMessage chat via the free Photon service, background sub-agents that spin up automatically with a live sub-agent tree, an Unreal Engine 5.8 MCP, a revamped desktop app, a dashboard profile builder, a browsable skills hub, smarter self-improving memory, and richer Telegram formatting.
Fahd Mirza installs a Pi-Reasoning fine-tune of Qwen 3.6 27B (Q4_K_M GGUF) with llama.cpp, wires it into Hermes agent, and tests it on bug-fixing, creative coding, and writing. The model uses baked-in multi-token prediction (MTP) plus speculative decoding for speed, fits in ~20GB of VRAM, and held up well on a real full-stack bug fix.
Fahd Mirza installs a coding-focused fine-tune of Gemma 4 12B — trained only on examples whose reasoning produced code that actually ran and passed tests — runs it locally with Ollama, and wires it into the Hermes agent. He puts it through three real tasks: fixing a tie-breaker bug in a World Cup tracker, generating an HTML animation from scratch, and optimizing a slow SQL query.
Bart Slodyczka installs Hermes Agent on a 16GB M4 Mac Mini and wires it to a local Gemma 4 12B model in LM Studio — using Claude Code to drive the whole setup. A practical local-first agent build.
Fahd Mirza pits Kimi K2.7 Code against GLM-5.2 inside Hermes Agent on a real buggy app — fixing a planted FIFA-rules bug, building a new bracket feature, and a one-shot HTML generation test.
Fahd Mirza runs Kimi K2.7 Code inside the Hermes agent — a 256k-context, always-thinking MoE model that refactored a 4,000-file SSL monitoring app and built five physics simulations from an image, all from single prompts.
Fahd Mirza drives the new open-source Nex-N2 (Next N2 Pro/Mini) through the Hermes coding agent on OpenRouter, building a full-stack Django + React banking app with intentional bugs to fix.
The Hermes Agent OS now includes an Obsidian Memory Galaxy that renders your entire Obsidian vault as a 3D star map. Hermes can query this galaxy for date-specific context, and every agent conversation automatically adds new notes back into the vault.
Hermes now includes an Idea Foundry pipeline where you submit an idea, agents classify and draft a plan, you approve at a single human gate, and sub-agents build the actual deliverable. Completed projects are stored in an Obsidian vault.
Google's Gemma 412B open model can be plugged into Hermes Agent OS as a free, local brain for full offline operation. A dual-brain setup lets a stronger model handle complex tasks while Gemma 4 handles lighter jobs. Requires 16GB VRAM or the free API.
Combining Claude with Hermes Agent gives Claude persistent memory across sessions via SQLite, an auto-improving skill system, and scheduled automation. Hermes can spawn Claude Code as a sub-agent, with Obsidian as a shared dashboard.
Instead of using Claude as a stateless chatbox, build an agentic operating system where Claude and Hermes share a single Obsidian memory vault in a command center dashboard with Kanban, goals, journal, and content studio.
Hermes V0.16 walkthrough: connect Gmail via MCP for 24/7 email management, enable real-time voice with MiniMax M3, use Jarvis for hands-free computer control, and run multi-agent Kanban with a built-in goal judge.
Hermes V0.16 adds a central dashboard for skills, schedules, and model switching; Jarvis voice agent for hands-free commands; free Nvidia Nemotron-3 Ultra and Step 3.7 Flash models; a one-click skill browser; and a no-terminal desktop app.
How to use the Hermes V0.16 Skills Hub Browser to find skills from skills.sh, Claw Hub, Claw Marketplace, and GitHub; preview skill.md files before installing; run automated security scans; and manage installed skills by trust tier.
Hermes can now be called on the phone using ElevenLabs for voice and Twilio for the phone number, with Hermes acting as the brain that has memory, tools, and past sessions. Claude 3.5 Haiku gave the fastest responses in testing.
Hermes Agent 0.16 adds a native desktop app for macOS/Linux/Windows, remote gateway support, a full web dashboard admin panel, fuzzy model picker, /undo command, and major skill cleanup.
Fahd Mirza runs JetBrains' new Mellum 2 model locally via vLLM, connects it to an MCP filesystem server for real file operations, and tests it inside Hermes Agent. Mellum 2 is Apache 2.0 licensed with a 131k token context window.
Tech With Tim's complete beginner guide to Hermes Agent — VPS setup, memory system (user.md, memory.mmd), 90+ built-in skills, soul personality file, crons, and the self-improvement loop that makes Hermes better over time.
Hermes shipped a full browser-based admin panel — manage models, sessions, skills, cron jobs, API keys, and MCP tools without ever touching a terminal. Covers the new chat interface, model switching for rate-limit avoidance, one-click skill install, and mobile responsive design.
Hermes Desktop is now in public preview on Windows, Mac, and Linux. This overview covers all 7 key features: real-time work visibility, drag-and-drop files, voice mode, easy settings, the command center, and multi-agent coordination.
How to install Claude Code and Hermes Agent side by side, wire them into a single workflow, and use new features: parallel dynamic workflows, skill bundles, Bitwarden key storage, prompt injection guard, and push notifications.
Side-by-side comparison of Hermes and OpenClaw based on real usage: which is smoother, which has better docs, unique features like Hermes Kanban and persistent goals, and when to use both together.
Fahd Mirza walks through installing and launching Hermes Desktop on Ubuntu, configuring Ollama as the local model provider, and exploring the new GUI's themes, personas, MCP tools, and session management.
Alex Finn gives a complete tour of the Hermes Desktop app — organizing sessions, the Artifacts panel for links and files, skills token savings, cron job management, multi-agent profiles, and memory compression tuning.
Answers to common Hermes questions: how to run multiple simultaneous tasks with the Kanban board, using persistent goals for autonomous work, and wiring Obsidian as external memory for long-term context.
Julian Goldie demonstrates a voice-controlled Hermes agent powered by MiniMax M3, letting you control your entire AI setup through real-time conversation. Supports multiple voice styles, runs locally, and connects to your full agent memory and tool stack.
FuturMinds walks through a complete setup for automating an entire social media operation with Hermes agent using the Zo MCP — covering post creation, comment reply automation, WhatsApp lead qualification, and daily analytics from Telegram.
Hermes now ships an MCP catalog — a curated list of reviewed tools you can install with a single command instead of manually editing config files. Includes per-tool action whitelisting, auto-updates, and parallel execution.
Fahd Mirza shows how DFlash's adaptive PFlash compression auto-tunes itself during Hermes agent sessions, cutting prefill from 3,572 tokens down to 148 — a 10× speedup with no manual config changes.
MiniMax M3, a new frontier-level agentic model from China, connects to Hermes agent via Ollama for free—giving Hermes a powerful brain for 12-hour autonomous tasks, local app control, web search, and scheduling.
Hermes Workspace is a community-built UI for Hermes agent that lets you add multiple AI models and launch agent swarms from one interface. This guide covers installation, OAuth model setup, and common sync issues.
Julian Goldie demonstrates how a custom Hermes Agent OS transforms scattered AI tools into a unified mission control dashboard with Kanban task boards, one-click MCP connections, and a shared Obsidian memory layer.
Combining Hermes agent with a custom Agent OS dashboard lets you access Claude, Gemini, and ChatGPT from one interface via OAuth—no separate API keys. One prompt triggers full multi-step workflows from research to publishing.
Three major Hermes updates: tool search auto-activates to boost accuracy from 49% to 74%, a built-in video agent powered by Hyperframes, and 30 days free of the Step 3.7 Flash model.
Julian Goldie answers community questions on Hermes: connecting Obsidian for personalized context, managing skills via Agent OS, creating named specialist profiles, and wiring Hermes into a Claude Code business dashboard.
Hermes v0.15 introduces automatic tool search that keeps only core tools in context and retrieves the rest on demand, raising tool-selection accuracy from 49% to 74% and enabling large toolsets without performance loss.
Wiring Hermes to an Obsidian vault creates a persistent memory layer where the agent already knows your goals, past tasks, and context on every run — and the vault is shared across Hermes, Claude, and OpenClaw.
Nous Research shipped a curated MCP catalog for Hermes that replaces manual config-file editing with an interactive picker. Tools are pre-reviewed, and you can allowlist or blocklist individual actions for fine-grained control.
Two free AI agents battle it out: OpenHuman offers instant no-code setup with auto-memory, while Hermes wins on customizability, MIT licensing, 200+ model support, and skill-based long-term memory.
Craig Hewitt, who runs two Hermes agents in production, debunks the six most common myths: you need a Mac mini, Hermes replaces Claude Code, build many agents, and more.
Hermes Agent v0.15 drops a velocity-focused release: 76% smaller codebase, sub-second startup, multi-agent Kanban swarms, 4500x faster search, and Bitwarden secret management.
Alex Finn's comprehensive Hermes Agent guide covers installation, model selection, when to use Hermes vs Claude Code vs Codex, /slashgoal, Kanban, self-improvement, and security best practices.
Alex Finn shares six Hermes Agent use cases from months of daily use: /slashgoal with metaprompting, Kanban board workflows, browser-based competitor research, and a personal memory wiki.
Julian Goldie demonstrates the Goldie Omnipresence Stack — a layered Hermes agent OS setup that auto-publishes SEO content, AI images, and videos across five websites simultaneously from a single keyword input.
Julian Goldie covers the Hermes update adding Grok 4.3's 1M-token context, pluggable video generation, /handoff model switching, 180x faster browser tool, Windows single-command install, and 22 messaging platforms.
Julian Goldie outlines 5 free Hermes configuration upgrades: Agent OS via soul.md, Kanban sub-agent teams, an AI SEO skill, hyperframe video templates, and goal-tracking with cron-based daily check-ins.
Alex Finn covers Hermes's 8 newest features: token-free session recall memory, /background multitasking, Grok 4.3 OAuth + X search, native Codex CLI integration, computer use, and AI video generation.
Full tutorial for deploying Hermes on a VPS to control WhatsApp groups through Telegram. Covers Hostinger one-click install, BotFather Telegram config, Periscope CRM MCP server setup, and automated briefings.
Owl Alpha on OpenRouter lets you run Hermes Agent at zero API cost. With a 1 million token context window and tool-use support, setup takes under 5 minutes via OpenRouter API key.
Alex Finn shows how to set up Hermes Agent backed by a local model on the NVIDIA DGX Spark, creating a private always-on AI employee. Hermes handles the full setup via a single prompt.
Nous Research added two free models to Hermes Agent: Qwen 3.6 Plus and Owl Alpha. Both run 24/7 tasks via the Nous Portal free tier with large context windows and no API costs.
Pareto Code from OpenRouter automatically picks the best coding model for each request. Combined with Hermes Agent, it routes tasks to whichever model scores highest — no manual model switching.
Hermes Agent (open-source planning framework) combined with AionUI's desktop interface creates a fully local agentic operating system — agents browse, write code, manage files, and chain in swarms autonomously.
Hermes Agent now supports 6 browser automation backends, including connecting to your real Chrome browser with existing logins and cookies via a single slash command.
Nate Herk walks through setting up Hermes Agent from zero on a private server — covering 91 built-in skills, voice output, scheduled crons, and a direct comparison with Claude Code and OpenClaw.
Hermes v0.13.0 ships a multi-agent Kanban board with hallucination gate, the /goal command for persistent task focus, session auto-resume, checkpoints v2, 8 security fixes, and new model support.
Alex Finn's comprehensive Hermes Agent setup guide covering reliability advantages over OpenClaw, Telegram + Opus configuration, self-improving skills, agent swarms, and local model options.
Hermes Desktop v0.6.0 ships a free Mac app with Kanban board support for the new multi-agent task system, file bookmarks for quick editing of skills and memories, and a searchable chat workbench replacing the terminal interface.
Alex Finn walks through the Hermes dashboard and Kanban board — run 10-30 simultaneous tasks with parallel worker agents, set up a librarian agent for admin work, and why Hermes now beats OpenClaw on reliability.
Hermes Agent v1.3 launches Curator, a feature that reads your task, picks the right agents and models automatically, chains their outputs, and learns from your feedback over time.
Hermes agent now integrates natively with LM Studio — local models are auto-discovered, loaded on demand, and Hermes picks the right reasoning level per model automatically.
Ollama's new 'ollama launch hermes' command sets up Hermes AI agent locally in seconds. Choose any local or cloud model, no config files, no API costs.
Anthropic's Claude Opus 4.7 pairs with Hermes, an open-source agent that writes reusable skill files after every task, creating an AI system that gets smarter the more you use it.
Craig Hewitt demos a Hermes + OpenClaw multi-agent architecture where Hermes acts as the always-on chief of staff and named OpenClaw sub-agents (Gary.
Step-by-step Hermes agent setup using fully local and free tools: Ollama for model hosting (Gemma 4 E4B), self-hosted Firecrawl for web search.
A walkthrough of a self-assembled Hermes dashboard that puts chat, a voice agent, a Kanban task board, model switching and an Obsidian-backed shared memory behind one screen.
A walkthrough of pairing Qwen 3.8 Max with Hermes as the agent harness — the model, harness and workflow-UI layer stack, Obsidian memory sync, and a lead-outreach workflow. No install steps or commands shown.
Julian Goldie describes wrapping Hermes in three layers — model, dashboard, and an Obsidian-based memory vault. Conceptual framing only; the build itself sits behind a paid community.
Julian Goldie tours a three-layer 'agent operating system' around Hermes — models, a dashboard replacing terminal use, and an Obsidian memory vault. A product tour of a paid-community build, not a setup guide.
Julian Goldie covers Moonshot's Kimi K3 open-weights release and running it behind Hermes Agent, plus a rundown of Hermes 0.19 'Quicksilver'. Largely a promotion for the presenter's paid community.
Julian Goldie demos five Hermes Agent workflows — news radar, lead outreach, video generation, an SEO pipeline, and a builder-judge review loop — run from a dashboard. Demo of a paid bundle, not a reproducible install.
Julian Goldie demos the open-source Hyperframes skill — now with keyframe and arc-motion support — paired with a Hermes-based video agent that scripts, generates B-roll, and assembles videos. Mostly a walkthrough of his paid Agent OS setup.
Tech With Tim explains why running an AI coding agent around the clock takes more than your own machine — local execution ties the agent to your computer staying awake and online — and points to a VPS for long-running and scheduled agent tasks.
Connect Hermes agent to LM Studio for fully local, free, private AI — no internet or subscription needed. Overview of the LM Studio + Hermes local setup.
The Hermes Skills Hub lets you search thousands of free skills from skills.sh, Claude Hub, Claude Marketplace, and GitHub, with one-click installation and built-in safety scanning.
Conceptual overview of building a custom 'agent operating system' around Hermes—unified memory via Obsidian vault, Kanban task boards, multi-model switching—instead of using the terminal directly. Analysis, not a setup guide.
Julian Goldie compares Odysseus (PewDiePie's new open-source AI agent) against Hermes Agent, finding Hermes better for multi-agent orchestration and 24/7 background operation while Odysseus suits single-agent local setups.
Quick overview of connecting Google NotebookLM to Hermes agent via MCP, letting Hermes automatically convert PDFs, docs, and videos into podcasts, infographics, and slide decks with a single command.
Quick overview of running Hermes agent free by connecting the Step 3.7 Flash model (free for 30 days) through an agent OS — Hermes handles autonomous tasks like file access and background work at no cost.
← Back to the full news digest · Browse the Hermes guides