amux vs Hermes Agent
Parallel coding agent fleet vs. a single self-improving personal AI assistant.
Hermes Agent by NousResearch is an open-source autonomous AI agent that runs as a persistent daemon, accumulates long-term memory, writes reusable skills from experience, and is accessible through messaging platforms like Telegram, Discord, and Slack. It launched in February 2026 and became one of the fastest-growing open-source AI projects with 175,000+ GitHub stars in four months.
amux is an open-source control plane for running an AI engineering team — launching, monitoring, and self-healing dozens of parallel Claude Code, Codex, or Gemini CLI sessions from a single web dashboard or mobile app. Built-in kanban board, scheduler, CRM, email, and browser automation. Single Python file, zero external dependencies, MIT licensed.
These two tools solve genuinely different problems. Understanding which one fits your situation is more useful than declaring a winner.
What each tool is optimized for
Hermes Agent is optimized for depth with one agent: persistent memory that survives across sessions, a self-improvement loop that writes reusable skills from experience, and ubiquitous access via 16+ messaging platforms. It gets smarter about you over time.
amux is optimized for breadth with many agents: running 10-50 coding agents in parallel, coordinating them via atomic task claiming so no two agents pick up the same work, and keeping them running overnight with a self-healing watchdog. It ships more code faster.
Feature comparison
| Feature | amux | Hermes Agent |
|---|---|---|
| Parallel agents | Unlimited (10–50+ typical) | Single persistent daemon |
| Primary interface | Web dashboard + mobile PWA + iOS app | Messaging platforms (Telegram, Discord, Slack, WhatsApp, Signal, 11 more) |
| Long-term memory | Per-session CLAUDE.md + notes | Core feature — cross-session memory with search |
| Self-improvement | No | Yes — writes skills from experience |
| Coding agents | Claude Code, Codex, Gemini CLI (full fleet management) | Code analysis tools (single agent) |
| Task coordination | Kanban board with atomic claiming (SQLite CAS) | Scheduled tasks (single queue) |
| Self-healing | Auto-compact, crash restart, YOLO mode, idle nudge | Not a focus |
| LLM support | Claude Code, Codex, Gemini CLI (provider routing) | Any LLM (Anthropic, OpenAI, Ollama, local) |
| Built-in tools | Kanban, CRM, Gmail, browser automation, scheduler, skills | 40+ tools (file ops, web search, terminal, code analysis) |
| Mobile | Native iOS app + PWA — manage fleet from phone | Messaging app access (WhatsApp, Telegram, etc.) |
| Cost | Free + your API tokens | Free + your API keys |
| License | MIT + Commons Clause | MIT |
| Hosting | Your machine (local-first) | Your infrastructure |
Hermes Agent's strengths
- Memory that compounds. Hermes remembers what you told it in January and applies it in July. amux agents start fresh each session (CLAUDE.md provides static context, but there's no cross-session experiential memory).
- Self-improvement loop. Hermes writes skills from what it learns and nudges itself to persist knowledge. Over time, it gets more capable at your specific tasks without you reconfiguring anything.
- Messaging-platform access. 16+ connectors mean you can reach your agent through whatever app you already use — no new dashboard to learn. If you live in WhatsApp, your agent lives in WhatsApp.
- Broad LLM support. Works with Ollama and local models, not just cloud APIs. Useful if you're running fully air-gapped or want to minimize API spend.
amux's strengths
- Parallel fleet at scale. Run 10 agents on 10 independent feature branches simultaneously. Hermes runs one agent; amux runs many — coordinated via a shared kanban board so tasks are claimed atomically and never duplicated.
- Coding-agent-specific tooling. Claude Code sessions, worktree isolation to prevent git conflicts, commit-guard, verify-gate, and a self-healing watchdog built specifically for how coding agents fail (context exhaustion, thinking-block corruption, tool-approval blocks).
- Overnight unattended operation. Queue tasks before bed. The watchdog handles all failure modes — auto-compact, crash restart, YOLO mode — and you wake up to completed pull requests.
- Full business ops stack. CRM, Gmail API email, browser automation, scheduler, skills, notes — all accessible to agents via the REST API. Hermes focuses on personal assistance; amux is a full business operations platform for AI-run teams.
When to use Hermes Agent
- You want a single intelligent assistant that knows you, remembers context, and improves over time
- You interact primarily via messaging apps (Telegram, Discord, Slack, WhatsApp)
- You want LLM flexibility including local models via Ollama
- Personal productivity is the goal — not shipping code at scale
When to use amux
- You want to run multiple coding agents in parallel on real codebases
- You need fleet coordination: shared task board, atomic claiming, no duplicate work
- You want agents running overnight, unattended, with self-healing
- You're managing an AI engineering team (solo or with others) and need a dashboard to see what every agent is doing
Can you use both?
Yes — they occupy different layers. Hermes Agent handles personal memory and communication; amux runs the parallel coding fleet that implements the work. The amux REST API (POST /api/sessions/NAME/send, GET /api/board) makes it easy for any external agent to delegate tasks to the fleet or check on progress.
Get started with amux
Run an AI engineering team from your dashboard or phone. Open source, Python 3 + tmux.
git clone https://github.com/mixpeek/amux && cd amux && ./install.sh
amux register myproject --dir ~/Dev/myproject --yolo
amux start myproject
amux serve # → https://localhost:8822