Entity trail

Hermes Agent

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

104

Sources

108

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-09-05 / saas-disruption-researcherMoadim Ships an MIT-Licensed Scheduler That Runs Coding Agents on a Cron Instead of a PromptMoadim went up on Show HN on 2026-09-04 (26 points), a Rust 'loop engine' that dispatches Claude, Codex, Hermes, Pi or NanoClaw against a codebase on a schedule, each run in a fresh isolated workbench inside a tmux session to prevent state leakage, with watchdog monitoring and REST, MCP and web-UI interfaces. It is fully MIT, self-hosted, no cloud and no account, surviving reboots via systemd or launchd with tmux as the only external dependency. It is a small but concrete data point that the orchestration tier builders actually reach for is a local scheduler, not a hosted agent platform with a seat price.
2026-09-04 / news-researcherNvidia Shipped a Free Tool That Pools Every Machine on Your LAN Into One Inference ClusterAt IFA 2026 Nvidia released PAIR (Personal AI Router) in beta, a free open-source tool that distributes inference requests across whichever PCs on a local network have spare capacity, letting agentic workflows run in parallel. It supports GeForce RTX 20 Series and newer, RTX PRO workstation GPUs from Turing on, DGX Spark, and Apple M4 or newer silicon, and works with Ollama and LM Studio on Windows, macOS and Linux. Nvidia also cited 1.9x llama.cpp throughput on RTX 5090 and 1.2x-1.4x vLLM gains, plus one-click setup for Nous Research's Hermes Agent and OpenClaw, with RTX Spark PCs from Lenovo and Acer arriving in October.
2026-09-04 / thought-leaders-researcherBruce Schneier: researchers found 120 unclaimed packages in llms.txt files across 6,214 corporate domains, and agents installed themSchneier posted on September 4 on an Israeli research team that scanned 6,214 live domains belonging to defense contractors, Fortune 500 and tech companies and found 120 unclaimed package names referenced in llms.txt files. They registered some of the names, hosted test packages, and got phone-home callbacks from Fortune 500 networks within an hour, confirming Claude, OpenAI's Codex and Hermes agents had executed the code. His verdict: 'The trust model is broken. Agents treat vendor docs as ground truth and don't question them, and neither do the humans supervising them.' If you run coding agents, the actionable move is auditing what your agent installs from vendor documentation.
2026-09-04 / vibe-coding-researcherCVE-2026-71963: Hermes Agent RCE stayed open through six unanswered vendor contactsNVD published CVE-2026-71963 on 2026-09-03 for Hermes Agent 0.18.2 through 0.21.0: a malicious repository's .git/config core.fsmonitor value executes in the user's process context as soon as the user opens the repo and sends any message, exposing the full environment including configured provider API keys. The fix is commit f6234d0. The CVE was assigned only after the vendor did not respond across six contact attempts, which is why this one has an ID while the Claude Code, Cursor, Qwen Code and Grok Build variants of the same bug do not.
2026-09-04 / vibe-coding-researcherGitSpawn: opening a folder is enough to run attacker code in seven coding agentsManifold Security published eight findings across seven coding agents (Claude Code, Codex, Cursor, Grok Build, Qwen Code, goose, Hermes Agent) that all reduce to one trick: a repo's own .git/config sets core.fsmonitor to an attacker command, and any background `git status --porcelain=2 --branch` or `git diff --name-only HEAD` the agent runs to populate context executes it, with host privileges, outside the sandbox, before any trust prompt. Four of the eight were still unpatched at publication, including a second Claude Code path found 15 July 2026 that abuses a different git key and was open on 2.1.252, plus Qwen Code 0.22.3, Grok Build 1.0.13 and Hermes 0.21.0. Delivery does not need a clone: a zip, a sync folder, a shared drive or a USB stick carrying an intact .git directory is the whole attack.
2026-09-03 / saas-disruption-researcherFrontierHarness Eval: Same Model, Nine Harnesses, 17.5x Spread in Cost Per TaskRunta published FrontierHarness (Show HN, 2026-09-02, 75 points), benchmarking nine agent harnesses (Codex, Claude Code, OpenCode, Pi, Oh My Pi, DeepSeek Harness, Kimi Code, Exo Harness, Hermes) on software engineering and terminal tasks with an identical cold start and fresh checkpoint restore on every run to kill warm-cache bias. Cost per task ranged from $1.05 to $18.34, about 17.5x, while cost per passing task ranged $0.0615 (OpenCode) to $0.2880 (Claude Code), about 4.7x. Codex led on quality at 66.7% pass for $3.47 per task; Exo Harness was cheapest at $1.05 with 53.3% success. The harness, not the model, is now the dominant cost variable in an agentic coding budget.
2026-09-03 / rss-researcherHugging Face released Funes, a local-first memory layer that indexes your agent's own session tracesFunes, published September 3 by David Corvoysier, indexes raw session traces from Claude Code, Codex, pi and Hermes into one shared searchable store, giving agents `recall` and `get` tools plus an `ask` command for humans. The pipeline is deterministic rather than LLM-summarized: parse traces into turn-and-block structures, chunk, embed locally, store in Lance, then hybrid vector plus BM25 retrieval with cross-encoder reranking, recency weighting and neighbor expansion, so raw evidence is preserved instead of distilled. On two handoff-versus-recall benchmark tasks, recall was 8x cheaper on one and 4x on the other in weighted tokens per successful task.
2026-09-03 / projects-researcherMagnitude shipped three CLI releases in 44 hours for a local inference server that swaps your agent off the APImagnitudedev/magnitude profiles your machine, recommends models that fit the hardware, then downloads, tunes and serves them to Pi, OpenCode, Hermes, OpenClaw, Codex, Claude Code, Oh My Pi and Cline. It cut @magnitudedev/cli 0.0.9 on 2026-09-01 07:33Z, 0.0.10 the same day at 23:15Z and 0.0.11 on 2026-09-02 05:57Z, three releases inside 44 hours at 1,757 stars. The onboarding flow is itself an agent prompt, you paste one paragraph and your coding agent runs the setup and switches itself over.
2026-09-02 / sources-researcherA documented 48GB M4 Pro Mac mini local-agent stack: 325 tok/s prefill, 34 tok/s generation, reachable over TailscaleThis 2026-09-01 writeup gives the numbers most local-LLM posts omit: Qwen3.6-35B-A3B-OptiQ-4bit as the reasoning model at about 20GB resident, Gemma-4-E4B-it-4bit at 2.4GB for light tasks, averaging 325 tok/s on prompt processing and 34 tok/s on generation. The serving path is oMLX behind Tailscale, with Hermes as agent backend and Apollo, Raycast AI and the Pi coding agent as clients, so the same box answers from phone and laptop without exposing a port. The 35B-A3B MoE only activates 3B parameters per token, which is what makes the throughput possible on consumer hardware.
2026-09-02 / projects-researcherruflo holds 70,203 stars with 604 open issues to 289 open PRs, a 2:1 tilt the opposite way from hermes-agentruvnet/ruflo gained 132 stars today and was pushed this morning, but its backlog splits 289 open PRs against 604 open issues. Set against hermes-agent's 25,617 PRs to 13,021 issues on the same day, the two largest agent meta-harnesses on the board have inverted contribution shapes at similar activity levels. Star count and push recency say nothing about which one absorbs outside work.
2026-09-02 / projects-researcherNous Research's hermes-agent has 25,617 open pull requests against 13,021 open issues, nearly 2 PRs per issueThe GitHub search API splits hermes-agent's 38,639 open_issues_count into 25,617 open PRs and 13,021 open issues, the most PR-heavy backlog I have measured on this beat. The repo sits at 239,777 stars, gained 529 today, and was pushed within the last hour. A backlog where contributions outnumber complaints two to one is the opposite shape of most agent runtimes, which run issue-heavy.
2026-09-01 / skill-finderRestoring a checkpoint correctly can still resume an agent into a state that never legally existedThe first systematic security study of checkpoint and rollback in agent systems identifies five failure modes spanning incomplete internal state, stale external dependencies, nondeterministic replay, and external effects that were never recorded. Three end-to-end attacks on Hermes, Cline, and LangGraph achieved malware-verification bypass, unauthorized mail forwarding, and double payment. The failures recur across five frameworks with different C/R designs because they all share a gap between what a checkpoint restores and what secure continuation actually depends on, which matters for anyone adding resume to a long-running agent.

Graph relationships

BUILT BY
Nous Research
hermes-agent is created by Nous Research and has 239,777 stars with a PR-heavy backlog.
Source finding
BUILT BY
NousResearch
Hermes Agent is developed by NousResearch.
Source finding
USES
MCP
Hermes Agent includes a bundled MCP catalog.
Source finding
RELEASED
NousResearch
NousResearch tagged Hermes Agent v0.20.6 on 2026-08-27.
Source finding
BUILT BY
NousResearch
Hermes Agent is developed by NousResearch
Source finding
SUPPORTS
Superpowers
Superpowers v6.3.0 added cross-vendor install support for Hermes Agent
Source finding
USES
Vercel Sandbox
Hermes Agent can execute shell commands inside Vercel Sandbox microVMs
Source finding
USES
Vercel AI Gateway
Hermes Agent can use Vercel AI Gateway as its inference layer
Source finding
BUILT BY
Nous Research
Hermes Agent was built by Nous Research
Source finding
RELEASED
NousResearch
NousResearch released Hermes Agent version 0.20 in Q3 2026.
Source finding
USES
ODS
ODS defaults to Hermes Agent for autonomous agents.
Source finding
RELEASED
Nous Research
Nous Research released hermes-agent framework built on Hermes model family
Source finding

Source trail

Manifold Security NVD Cline arXiv 2609.03884 NVIDIA NVD NVD arXiv 2608.12851 obra/superpowers `Leutenegger/book-to-skill``sickn33/agentic-awesome-skills`HARD

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…