How I Turned Hermes into My $90/Month Investment Analyst

TL;DR

After a month of trial-and-error, 0xJeff has configured Hermes Agent (Nous Research's open-source AI agent) to act as a full-stack investment analyst for roughly $90/month. The setup covers daily macro and tech briefings via the X API, portfolio tracking across crypto and equities, personalized investment research grounded in persistent memory, and position exit alerts. The moat is context - not raw model intelligence.

Why Discord Wasn't Enough

Most Hermes users start on Discord. It works - but you hit a wall fast. Discord can't render HTML, show charts, or give you anything close to the visual feedback loop you get from Claude Artifacts or ChatGPT's canvas.

The unlock was OpenWebUI. Connecting Hermes to OpenWebUI transforms it from a text terminal into a full visual workspace. You can ask Hermes to design an HTML explainer, a comparison table, or a dashboard layout - and it renders right there in the chat. Same artifact-style experience as Claude, at a fraction of the price.

Hermes Workspace also exists - with Swarm (multi-agent orchestration), editable Skills, Tasks, Memory, and Soul files all in one interface. Think of it as Hermes command center. It's powerful, and there's more to explore there.

What Hermes Does Every Morning

The daily briefing system is the most valuable part of this setup. Via the X API ($10-15/month), Hermes tracks a curated list of accounts and delivers automated reports every morning to Discord:

• Tech Report - synthesizes tweets from accounts like @SemiAnalysis_, filtered for what matters to the portfolio
• Macro Report - pulls from @KobeissiLetter and major news sources
• Bookmark Briefing - queries the last 24 hours of X bookmarks, scores each by relevance, and ranks them so you read the highest-signal items first
• Top 5 Synthesis - Hermes reflects across all the day's reports and picks the five most important insights with explanations of why they matter
• Polybond Morning Brief - surfaces sharp signals and potential insider activity from a personal Polymarket dashboard

The result: zero manual scrolling. Open Discord, read the summary, done.

The Memory Moat

This is where Hermes actually pulls ahead of a standard LLM.

A normal GPT or Claude session forgets everything the moment it ends. Hermes uses a multi-layer memory architecture - SQLite databases with full-text search, memory.md files, and Honcho persistent user profiles - to retain context across every session. It builds a durable model of who you are as an investor.

In practice, this means Hermes always knows your:

• 3 core investment theses
• Risk appetite and position sizing rules
• Current holdings across crypto, equities, and prediction markets
• The rationale behind each trade

When new research comes in, Hermes filters and frames it through your specific context. It doesn't just surface information - it evaluates whether that information is relevant to your portfolio and your strategy.

The memory loop runs like this: daily cron jobs ingest new insights, Hindsight processes and synthesizes patterns across any time window, and Hermes draws relationships between things you discussed weeks ago. Normal LLMs can't touch that. The moat is in the data and context, not raw intelligence.

The Full Cost Breakdown

Here's the exact monthly spend to run Hermes as a comprehensive investment analyst:

Total: ~$90-95/month for what would cost thousands per month with a human analyst team, or $100+/month just for Claude Max if you tried to build the same thing with Claude alone.

DeepSeek v4 Pro currently has a 75% discount running through the end of May 2026. DeepSeek v4 Flash handles simpler tasks even cheaper. If the discount ends, Opencode Go ($5 for the first month) is a solid fallback entry point.

What Hermes Won't Do

One deliberate constraint: Hermes is not allowed to execute trades directly.

This isn't a technical limitation - it's a risk management decision. AI hallucinations are real, and an agent that confidently buys the wrong ticker at 3x leverage is a nightmare scenario. The current setup uses Hermes for research and alerts only, keeping a human in the loop for all execution.

What Hermes does do is track positions and send alerts when they hit price targets. This turns out to be more valuable than it sounds - it solves the bag-holding problem. Without alerts, the temptation is to stay in a position too long, average down on a losing trade, and slowly become a full-time community member in the wrong direction. Hermes exits you on time.

Other current limitations worth knowing:

• Cron jobs run in fresh sessions with no memory carryover - each scheduled prompt must be fully self-contained
• Running locally on Ollama requires at least 24GB VRAM for adequate context (16k-32k tokens minimum for agent use)
• Token costs can escalate quickly without proper configuration - Hermes uses dual compression and Anthropic prompt caching to manage this

Where This Goes Next

Hermes Agent is at v0.12.0 with 32,000+ GitHub stars and growing fast. The agent framework space is splitting into two camps: API-based agents vs local model advocates. Hermes is the flagship for the local/cheap-inference camp, and the community momentum reflects that.

For 0xJeff, the next steps include:

• Deeper exploration of Hermes Workspace and Swarm (multi-agent missions)
• Expanding equities coverage as positions diversify beyond crypto
• A dedicated resource page for Hermes setup guides and workflows

The setup is still evolving - "first Frieza form" in the author's words. But the core loop is working: consistent daily signals, personalized research, and position discipline. For anyone spending hours manually scrolling through X or paying for expensive research subscriptions, it's worth a serious look.

Sources: Nous Research Hermes Agent, DataCamp Tutorial, Daily Briefing Bot Guide.