Llmquant Hermes

llmquant-hermes

English | 简体中文

Turn your Hermes agent into a financial analyst. Ask it anything from equities to prediction markets — and let it keep watch: morning briefs, earnings pings, 13F diffs, delivered while you sleep. Powered by LLMQuant Data.

What you get

One paste gives your Hermes agent an analyst's full stack — market data, analysis method, and five standing watchers it runs for you:

• Morning brief — a pre-open rundown of macro, your watchlist's overnight moves, fresh company news, earnings-adjacent filings — and overnight prediction-market odds moves, cross-checked against the news — waiting when you wake up.
• Earnings watch — a nudge when a name on your list reports or updates guidance, so you never miss a print.
• 13F diff — what the big managers you track bought, sold, or trimmed in their latest 13F, with news context on the changed names.
• Portfolio pulse — a weekly check-up on your saved holdings: trend, macro, and news that touches what you own.
• Polymarket watch — a short morning list of prediction markets worth a look: the biggest overnight odds moves, fresh high-volume markets, and where the odds disagree with the news.

Alongside them come the 18 LLMQuant/skills analysis skills (/llmquant-equities, /llmquant-macro, …) for on-demand deep dives — five standing watchers plus a full analysis toolkit. Type /llmquant in Hermes and all of it shows up:

Contents

• [What you get](#what-you-get)
• [Setup](#setup)
• [Use it](#use-it)
• [What it looks like](#what-it-looks-like)
• [FAQ](#faq)
• [Reference](#reference)
• [Contributing](#contributing)
• [License](#license)

Setup

You need a running Hermes agent, a messaging channel (Telegram, Discord, Slack, …), and a model backend. These playbooks are tuned for a cost-conscious DeepSeek baseline.

If you don't have Hermes yet

Hermes Agent is a separate project from Nous Research. Install it first:

• Linux / macOS / WSL2: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
• Windows (PowerShell): iex (irm https://hermes-agent.nousresearch.com/install.ps1)
• Desktop GUI (Electron, macOS / Windows): run hermes desktop after the CLI install, or grab the installer from the official site.

Then run hermes model to pick a provider and hermes to start the terminal TUI. Full docs: getting started · NousResearch/hermes-agent. From here on this README assumes Hermes is running.

If you already have Hermes

Paste this into your agent — it is the whole setup:

Read https://raw.githubusercontent.com/LLMQuant/llmquant-hermes/master/SETUP.md and set up LLMQuant Data for me. Use DeepSeek as the baseline model. Only ask me for things you can't do yourself.

Your Hermes agent will:

• connect the hosted LLMQuant Data MCP URL first, with local stdio as the fallback;
• confirm your Hermes model, with DeepSeek as the recommended baseline;
• install the five market playbooks, the analyst persona, and the LLMQuant/skills analysis skills;
• smoke-test the connection with a 0-credit call;
• get you to a first answer right away — a question to try or a manual morning-brief run (scheduling is optional, later);
• ask before it edits config, writes cron jobs, or needs an API key.

If the raw URL is unreachable, point the agent at the local [SETUP.md](SETUP.md) — the same runbook.

Prefer to wire it up by hand?

1. Sign up at llmquantdata.com (Free Beta is enough to start), then copy a hosted Remote MCP URL from Dashboard -> Connect.
2. Add one entry to ~/.hermes/config.yaml, then reload MCP (or restart Hermes):

``yaml mcp_servers: llmquant-data: url: "https://mcp.llmquantdata.com/u//mcp" ``

1. Call personal_profile once — a 0-credit auth smoke test.
2. Install the skills:

``bash hermes skills tap add LLMQuant/llmquant-hermes hermes skills tap add LLMQuant/skills hermes skills list ``

Tap may register a source without installing it (see the [FAQ](#faq)). If skills are missing from the list, install them with your Hermes version's skill flow, or clone each repo and copy its skills/ into ~/.hermes/skills/.

1. Copy [templates/SOUL.md](templates/SOUL.md) to ~/.hermes/SOUL.md, and [templates/AGENTS.md](templates/AGENTS.md) into the workspace that will run the market jobs (edit the watchlist and delivery settings).
2. See it work now: ask Hermes a question or run one playbook manually. Scheduling is optional — see [Use it](#use-it).

Local stdio still works — use [config/mcp-stdio.example.yaml](config/mcp-stdio.example.yaml) and edit the YAML by hand (see the [FAQ](#faq) on the hermes mcp add --env bug).

Use it

You don't need cron to get value. Once setup finishes, just talk to Hermes.

Ask in plain language — Hermes picks the right LLMQuant Data tools:

• How do you see NVIDIA's price action these past few days?
• What changed in Berkshire's latest 13F?
• Any notable news on my watchlist today?
• Which prediction markets moved overnight — and does the news back the move?

Run a playbook with a slash command. Every installed skill is a /skill-name you can follow with a task:

• [/morning-brief](skills/morning-brief/SKILL.md) — the full pre-open brief, on demand.
• [/earnings-watch](skills/earnings-watch/SKILL.md) — scan your watchlist for earnings and guidance.
• [/13f-diff](skills/13f-diff/SKILL.md) compare my tracked managers to the last run — pass an inline instruction.
• [/portfolio-pulse](skills/portfolio-pulse/SKILL.md) — a check-up on your saved holdings.
• [/polymarket-watch](skills/polymarket-watch/SKILL.md) — this morning's interesting prediction markets, on demand.

Go deeper with the analysis skills. The LLMQuant/skills toolkit triggers the same way — /llmquant-equities, /llmquant-macro, /llmquant-prediction-markets, 18 in all; run hermes skills list to see them. You can stack up to five skills in one message.

Want a playbook to run on its own? Each SKILL.md carries a cron command. hermes cron trigger fires the job immediately, so you can confirm it works without waiting for the next scheduled run.

What it looks like

One paste sets everything up. The agent connects the MCP, installs the skills, and hands you first steps:

Ask in plain language. NVDA's price action, answered from LLMQuant Data tools with the week's news behind it:

It knows your saved holdings. personal_holdings turns your LLMQuant profile into a portfolio snapshot (demo profile shown — illustrative data):

FAQ

Remote MCP URL or local stdio? Prefer the hosted Remote MCP URL — it needs no local Node.js and avoids a stdio bug where hermes mcp add --env wrote the key into args instead of env (seen on v0.18.0). If you do use stdio, put LLMQUANT_API_KEY under env and edit the YAML by hand.

hermes skills tap add didn't install anything? Tap can register a source without installing it. Run hermes skills list; if a skill is missing, install it with your Hermes version's skill flow, or copy the repo's skills/ into ~/.hermes/skills/ — Hermes only reads that global path.

My cron fired at the wrong time? Cron runs in the host machine's local time on some Hermes versions. Confirm the next-run time with hermes cron list after scheduling; every 3d parses on v0.18.x and later.

What does it cost? Two separate bills. Hermes model inference is paid to your model backend (Nous Portal, DeepSeek, OpenRouter, …); LLMQuant credits are spent only when a playbook calls LLMQuant Data tools. The Free Beta plan is enough to try things out — current plans are at llmquantdata.com.

Reference

How this fits together

Three layers, wired by one paste: LLMQuant Data is the tools (prices, filings, 13F holdings, macro, and company news over MCP), LLMQuant/skills is the analysis method, and Hermes is the runtime (cron, messaging, memory). This repo is the glue — it answers how to watch, not how to analyze: when a job runs, which tools it calls, where the output goes. Alongside the playbooks it ships the setup runbook ([SETUP.md](SETUP.md)), an analyst persona and workspace templates ([templates/](templates/)), and MCP config examples ([config/](config/)).

Company news

The playbooks lean on news_browse, the LLMQuant Data tool for recent company news. It returns an AI-written title, abstract, summary, event/topic labels, tickers, publication date, and a source link. It is exact filtering, not semantic search — query by ticker, event, topic, date range, and limit. Coverage runs from April 11, 2026 onward.

Repository layout

llmquant-hermes/
├── README.md
├── README.zh-CN.md
├── SETUP.md
├── LICENSE
├── docs/
│   └── images/            # screenshots referenced by "What it looks like"
├── skills/
│   ├── llmquant-setup/SKILL.md
│   ├── morning-brief/SKILL.md
│   ├── earnings-watch/SKILL.md
│   ├── 13f-diff/SKILL.md
│   ├── portfolio-pulse/SKILL.md
│   └── polymarket-watch/SKILL.md
├── templates/
│   ├── SOUL.md
│   └── AGENTS.md
└── config/
    ├── mcp-remote.example.yaml
    └── mcp-stdio.example.yaml

Contributing

Keep this repo small: add a playbook only if it needs Hermes-specific capabilities — cron, messaging, persistent memory, or standing agent context. Anything that works equally in Claude Code, Cursor, or Codex belongs in LLMQuant/skills, not here.

License

MIT. See [LICENSE](LICENSE).

LLMQuant

Open-source community for AI, LLMs, and quantitative finance.

LLMQuant Data · Data MCP · Skills

Website · GitHub · LinkedIn

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: LLMQuant
• Source: LLMQuant/llmquant-hermes
• License: MIT
• Homepage: https://llmquantdata.com

Install and usage instructions live in the source repository linked above.