hermes-agent.nousresearch.com via Reddit

Nous Research Ships Hermes Desktop Across Mac, Windows, Linux

agents open source ai-agents open-source

Key insights

  • Hermes Desktop v0.15.2 ships for macOS 12+, Windows 10/11, and Linux under the MIT license with persistent memory that auto-generates skills across sessions.
  • All Nous Portal paid tiers include monthly credits and access to over 300 models with built-in tool use.
  • Five execution backends (local, Docker, SSH, Singularity, Modal) support deployment with container hardening and namespace isolation.

Why this matters

An open-source, MIT-licensed agent that persists skills and memory across Telegram, Discord, Slack, WhatsApp, Signal, email, and CLI changes the economics of deploying AI assistants for small teams: no vendor lock-in, no per-session context loss. The five-backend execution model spanning local through Modal and Singularity means the same agent can serve air-gapped HPC environments and scalable cloud deployments alike, broadening the viable deployment range well beyond typical commercial agent offerings. For AI practitioners evaluating open-source agent frameworks, Hermes Desktop sets a concrete benchmark for cross-channel, persistent-memory agent infrastructure without relying on proprietary APIs.

Summary

Nous Research has shipped Hermes Desktop, a native app for macOS 12+, Windows 10/11, and Linux running Hermes Agent v0.15.2 under the MIT license. The app builds persistent project memory, auto-generates reusable skills, and retains past solutions across sessions. It spans Telegram, Discord, Slack, WhatsApp, Signal, email, and CLI -- one agent, one memory layer on every surface. Five execution backends (local, Docker, SSH, Singularity, Modal) include container hardening and namespace isolation. Essentially: (Nous Research) ships an agent that accumulates project knowledge rather than resetting per session. - Paid Nous Portal tiers (Plus, Super, Ultra) unlock 300+ models with built-in tool use. - Isolated subagents each get their own terminal and Python RPC scripts for zero-context-cost pipelines. - Natural-language scheduling handles reports, backups, and briefings running unattended through the gateway. The architecture targets teams who need full infrastructure control without giving up persistent AI context.

Potential risks and opportunities

Risks

  • Nous Portal's model access layer (300+ models) creates a hosted-infrastructure dependency, meaning teams running production workloads through Hermes Desktop are exposed if Nous Research changes pricing or availability.
  • Multi-channel memory persistence across WhatsApp, Signal, and Telegram raises data residency and compliance exposure for enterprise users in regulated sectors, where agent-held conversation history can trigger audit obligations.
  • The MIT license enables forks and competing distributions, which could fragment the skill and memory ecosystem and reduce the quality of any shared knowledge base over time.

Opportunities

  • HPC and research computing teams (universities, national labs) gain a ready-made agent layer spanning Singularity and SSH backends, removing a standard integration barrier for AI tooling in those environments.
  • Messaging automation developers can build on Hermes Desktop's channel layer (Telegram, Discord, Slack, WhatsApp, Signal) to deliver persistent-memory AI features without constructing agent infrastructure from scratch.
  • Nous Research can monetize the open-source agent through the Nous Portal's tiered credit model (Plus, Super, Ultra) while the MIT license drives community-contributed skill library growth.

What we don't know yet

  • Pricing for the Plus, Super, and Ultra Nous Portal tiers is not disclosed on the release page.
  • The monthly credit amounts included with paid portal tiers are not specified in the release material.
  • No roadmap is given for additional channel integrations beyond the current seven (Telegram, Discord, Slack, WhatsApp, Signal, email, CLI).