TODO for AI
vs
Hermes Agent
Open-source self-improving personal agent from Nous Research: skills it writes itself, memory that persists, reachable from 20+ chat apps. Here is where each one wins, with numbers, and who should pick which.
Hermes Agent
TODO for AIWho should pick which
- You want MIT source and will host it on your own box or VPS
- You mostly talk to your agent from Telegram or Discord
- You want an agent that writes its own skills over time
- You want a list the agent works through, not a chat you scroll
- You want the machine, browser and vault managed, with your PC attached when needed
- You want one flat plan across Claude, GPT, Gemini and Grok
- You need team roles, pooled usage and per-tool approvals
What Hermes Agent is, and where it stops
Hermes Agent is Nous Research's open-source personal agent and the current centre of gravity for the self-hosted crowd. It is MIT licensed, runs on a $5 VPS or your laptop, and talks to you from Telegram, Discord, Slack, WhatsApp, Signal, email or the terminal. What sets it apart is the learning loop: it writes skills from tasks it has done, refines them on the next run, and keeps a layered memory of you across sessions. A cron scheduler and subagents cover unattended and parallel work, and it can drive a browser. It is positioned as the successor to OpenClaw and imports OpenClaw settings, skills and memories with one command.
The trade is the same as OpenClaw's. You are the operator: install, ports, updates, sandbox choice, and each provider bills you separately unless you buy Nous Portal credits. There is no hosted machine, so it is only as always-on as the box you keep running, and no cloud browser. Work lives in chat threads rather than a list, so nothing tracks what is open, done or waiting on you.
Technical users who want a self-hosted agent that learns them over time and lives in Telegram, Discord or the terminal.
Three reasons Hermes Agent users look elsewhere
Gateway, sandboxing, updates and key management are yours, on a machine you keep up. The install is one line; the operating is not.
Tasks are messages in a chat. Nothing shows what is queued, running, blocked on you or finished, and nothing proposes the next one.
It can browse, but not as you in the Chrome profile where you are already logged in to your CRM, ad account or bank.
Six dimensions that decide it
What each one ships
Same ladder, different meter
Portal plans are credit pools, 200+ models. Self-hosted is per-token per provider.
One pool priced in dollars of model cost, spendable on any vendor. Session and weekly windows pace it, and a 1.3-1.5x plan multiplier applies, so a $20 plan is about $15 of list-price API. Team seat prices go into one shared pool.