Hermes Agent

Hermes Agent is Nous Research's open-source (MIT license, verified via the GitHub API) personal AI agent. It connects to Telegram, Discord, Slack, WhatsApp, Signal, email and a CLI as messaging surfaces, plus native desktop apps for macOS and Windows and a one-line curl installer for Linux/macOS (curl -fsSL ^[6] | bash). The install script git-clones NousResearch/hermes-agent's main branch and does an editable pip install rather than pulling from PyPI, which matters for the security point below.

The agent writes its own reusable skill files from solved tasks, keeps persistent memory across sessions, supports natural-language scheduling for recurring jobs, and can delegate to isolated subagents with their own conversations and terminals. Code and command execution run through one of five pluggable sandbox backends (local, Docker, SSH, Singularity, or Modal) with container hardening and namespace isolation, per the vendor's own SECURITY.md, which is an unusually detailed and candid trust-model document (it names the one boundary it treats as load-bearing and states plainly that other layers are not).

GitHub shows 396 contributors and near-daily tagged releases (e.g. v0.20.4 on 18 August 2026 rolled up roughly 74 merged PRs since the prior patch); top contributor is Nous co-founder teknium1 with 8,644 contributions. Parent company Nous Research (founded 2023) had raised about $70M from Paradigm, Robot Ventures, North Island Ventures, Delphi Ventures, OSS Capital and Balaji Srinivasan, and was in talks for a further $75M Series B at a $1.5B valuation led by Robot Ventures as of TechCrunch's July 13, 2026 report — that round's close status is not established as of this listing.

The product page shows four tiers (Free, Plus, Super, Ultra) sold through a separate "Nous Portal," described as monthly credits toward 300+ models plus built-in tool use, with no numeric prices on the Hermes Agent site itself. The portal page that would list exact figures (portal.nousresearch.com) returned an HTTP 429 "Vercel Security Checkpoint" page on every attempt today — direct curl, a proxy fetch, and a full headless browser all hit the same block — so current tier prices are not established from a primary source. TechCrunch's July 13, 2026 reporting put paid tiers at "$20-$200/month," which is the only independently dated figure available; treat it as a rough range that may have since changed rather than today's price.

osv.dev lists 10 distinct CVEs against hermes-agent published between May and August 2026 (CWE categories mostly injection, plus one DNS-rebinding bypass on WebSocket endpoints rated HIGH, and one world-readable-secrets-file bug rated MODERATE). All were publicly disclosed by one external researcher; each GitHub advisory states "the vendor was contacted early about this disclosure but did not respond in any way," so these were not privately coordinated. Nine of the ten have merged fixes on GitHub (versions 0.11.0 through 0.18.0). The tenth, CVE-2026-10224 (moderate-severity resource exhaustion via the Feishu webhook handler), has a fix PR (#73406) that is still open and unmerged as of 20 August 2026. Separately, the PyPI package is one release behind GitHub (0.19.0, uploaded 20 July 2026, versus the current v0.20.4 tag) and predates a merged injection fix (CVE-2026-10221, merged 23 July 2026) — the official installer avoids this by cloning the main branch directly rather than installing from PyPI.

GitHub's issue search shows roughly 11,400 open and 11,400 closed issues (type:issue, excluding PRs) for a repo created in July 2025 — a large raw volume in either direction for a 13-month-old project, and roughly half of everything ever filed is still open. No independent reviews turned up on Reddit, Trustpilot, or G2 as of this search; that's normal for a young, fast-moving open-source project rather than a red flag on its own.