Hermes Agent

Free

Hermes Agent is an open-source AI agent you can host yourself. It connects to your Telegram, Discord, Slack, WhatsApp, Signal, or email so you can talk to one agent across all your apps.

The agent remembers your projects, writes new skills for itself, and can run scheduled tasks in the background. You control the infrastructure, the models, and your data.

This is AI agent work for people who build and run their own systems. It is not a plug-and-play tool.

Open it if you want a learning agent you own end-to-end. Skip it if you want something that just works today.

These are the people who get the most from the time they put in.

• Users wanting model flexibility without lock-in: You switch providers with one command. You are not stuck with one API.
• People who want agent access across multiple messaging apps: You use Telegram in the morning and Slack at work. Same brain, both places.
• Users prioritizing data control over convenience: Your memory files live on your server. No cloud AI service holds your context.
• Developers who want agents that optimize their own workflows over time: The agent writes skill files after you complete a task. The next run is faster.
• Solo developers and researchers using agents daily over months: It learns how you like outputs formatted. You stop repeating yourself.
• Teams who want cron-scheduled autonomous tasks with safety rails: You set "every morning at 8" in plain language. It fails closed, not loud.
• Multi-channel operators needing one brain across Slack, Discord, Telegram: One persistent context follows you. No fragmentation.
• Solo builders who want an agent live tonight: The hosted version deploys in one click. You skip the server setup.
• People running overnight research tasks where latency matters less than completeness: It works while you sleep. You check results in the morning.

These are what you notice after a week of real use.

• Persistent memory across sessions: It loads MEMORY.md and USER.md at startup. You do not start from zero each chat.
• Self-written skills that compound: After you walk through a task a few times, it drafts a skill file. Future runs need less guidance.
• One agent across 20+ platforms: Telegram, Discord, Slack, WhatsApp, Signal, email, and CLI all talk to the same brain.
• Scale-to-zero pricing on the hosted version: You pay only when the agent is working. Idle time costs nothing.
• Isolated subagents for delegation: You spawn separated agents with their own memory and tools. Parallel work, no context window bloat.
• Multiple sandbox backends: Local, Docker, SSH, Singularity, or Modal. You pick the isolation level you need.

These are the real costs you pay for the control you get.

• Setup is not casual: You flip config switches for persistent memory and skill generation. Both are off by default. You might miss them and wonder why it is not learning.
• Self-hosted means ongoing maintenance: You handle reboots, supervision, and infrastructure. There is no one to call.
• Memory caps are small: The identity file tops out around 3,500 characters total. You will hit limits and need to consolidate manually.
• Windows users report real problems: Antivirus false positives on core files. Uninstallation can leave stubborn processes running.
• Token overhead is high: Expect roughly 73% more tokens per API call. That adds up fast.
• Not built for coding: Cursor or Claude Code will be faster for pure software development.