Product / UX / Orchestration & agents

Hermes Agent

Self-improving autonomous agent from Nous Research that grows through persistent memory, auto-generated reusable skills and a built-in learning loop. It runs on a VPS or a serverless backend, is driven from Telegram and other messaging surfaces, and ships subagent delegation, natural-language scheduling, sandboxing and more than forty tools.

Verified 2026-08-13 via the NousResearch/hermes-agent repository and the Hermes Agent product page.

Openness

5 high confidence

5.0

license: MIT(OSI)
source: public(github.com/NousResearch/hermes-agent)
core-gated: ungated

Fully MIT-licensed, complete public source, self-hosted by design, no proprietary core; top openness.

https://github.com/NousResearch/hermes-agent recorded 2026-08-13
MIT license badge and README statement "License - MIT, see LICENSE"; full public Python source for the self-improving multi-channel agent; 230.0k stars, 45.5k forks
https://hermes-agent.nousresearch.com recorded 2026-08-13
product page labels the desktop download "Open Source - MIT License" and footers "Hermes Agent v0.20.0, Nous Research, MIT License - 2026"

Adoption

4 medium confidence

4.0

A real usage signal beyond stars: Hermes Agent is reported #1 on OpenRouter's global token rankings, confirmed by Nous Research's own announcement of roughly 271B tokens, with secondary reports giving about 224B tokens a day and putting it ahead of OpenClaw's roughly 186B. The OpenRouter app page states 35.8T total tokens, a daily global rank of #1, 417 models used, and #1 in both Productivity and Coding Agents. GitHub stars, about 181K in under four months, corroborate but are not the basis. What is banded is inference token volume, which is neither a download count nor a user count, so reach is inferred from tokens rather than a disclosed installed base - hence medium confidence.

https://github.com/NousResearch/hermes-agent recorded 2026-08-13
230.0k stars, 45.5k forks, active releases (corroborating, not the basis)
https://x.com/NousResearch/status/2052904761087729897 recorded 2026-08-13
PRIMARY - Nous Research: "Hermes Agent is now #1 on the Global @OpenRouter token rankings."
https://openrouter.ai/apps/hermes-agent recorded 2026-08-13
PRIMARY - OpenRouter app page reports total tokens 35.8T, daily global rank #1, active since Mar 2026, 417 models used, #1 in Productivity and #1 in Coding Agents
https://chatforest.com/builders-log/hermes-agent-nous-research-openrouter-number-one-self-improving-builder-guide/ recorded 2026-08-13
aggregator detail on the MIT self-hosted agent and its star growth (corroborating)

Capability

4 medium confidence

4.0

Distinctive autonomy (self-improvement loop + persistent memory + subagent delegation) and very broad tool/channel coverage place it high on the feature matrix; no published benchmark so scored 4, just below a verified-benchmark ceiling.

https://github.com/NousResearch/hermes-agent recorded 2026-08-13
self-improving skills, persistent memory, 40+ tools, subagents, natural-language cron and sandboxing; no benchmark figure published
https://openrouter.ai/apps/hermes-agent recorded 2026-08-13
independent description - runs persistently with memory across sessions, with web search, browser automation and vision, plus scheduled automations and subagents

Verified 2026-08-13