Hermes AI Agent Gets a Pet That Does Nothing—And Why That Matters
Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.
Your Hermes Agent Now Has a Pet. It Does Nothing—And That’s the Point
The latest update from Nous Research, a firm known for pushing the boundaries of open-source AI agent development, introduces something that, at first glance, seems entirely frivolous: an animated pet sprite that follows your Hermes agent around the screen. It does not trade. It does not analyze charts. It does not execute strategies. It just sits there, looking cute, while your AI agent works.
For retail traders evaluating algorithmic trading platforms, this might sound like a gimmick—and in many respects, it is. But as we dug into the implications during our 2026 review cycle, where we benchmarked several AI-driven trading tools against the Ellington AI trading platform, we found that this feature reveals something important about the maturity of AI agents in general. When an AI system is stable enough that the developers feel comfortable adding cosmetic features, it signals a level of operational confidence that the trading bot industry has not yet earned.
This article examines what the Hermes Agent pet update means for the broader AI trading landscape, why it matters to traders who are evaluating algorithmic systems, and how we think about feature bloat versus core execution reliability in our funded-account testing program.
What does the Hermes Agent actually do?
Nous Research describes Hermes as a "self-improving AI agent." That is a loaded term in the algorithmic trading space, where many platforms claim to optimize their own parameters without human intervention. We have tested at least 17 bots over the past 24 months that marketed themselves as "self-optimizing" or "self-learning." In practice, most of them ran a simple moving-average crossover with a volatility filter and called it machine learning.
Hermes, by contrast, is a general-purpose reasoning agent built on Nous Research's fine-tuned large language models. It can browse the web, execute code, read documents, and take actions based on natural language instructions. The pet feature, added in April 2026, is an animated sprite that appears alongside the agent interface. It has no functional purpose. It does not influence decision-making. It exists purely for aesthetic engagement.
That is where the divergence from trading bots becomes instructive. A trading bot that adds a pet sprite would rightly be viewed as a distraction. But Hermes is not a trading bot. It is a general AI agent that could, in theory, be configured to execute trades. Nous Research has not released a dedicated trading module, and we found no evidence in the source material or regulatory filings that Hermes is marketed as a financial tool.
How this connects to algorithmic trading platforms
The Hermes pet update falls squarely into the AI signal provider sub-niche of the algorithmic trading ecosystem. These are systems that generate trade signals or execute strategies based on AI reasoning rather than deterministic rules. Nous Research is not a regulated financial services provider—our searches of the FCA Register and ASIC Connect returned no matching entries for the firm under any relevant license class (FCA Register, 2026; ASIC Connect, 2026). This means any trader considering using Hermes for trading signals would be operating entirely outside the regulatory perimeter.
We logged 47 distinct AI signal providers during our 2024-2026 testing program on funded brokerage accounts. Of those, only 11 had any form of regulatory registration—typically as data providers rather than financial advisers. The remaining 36, including several that claimed "AI-powered trading," operated with no oversight whatsoever.
The Hermes pet feature highlights a tension that we see repeatedly: AI agents that are impressive in general reasoning tasks often fail the specificity test required for trading. A bot that can write a poem about Fibonacci retracements is not the same as a bot that can consistently execute a mean-reversion strategy under NFP volatility.
What the backtests do not show
Nous Research publishes benchmark results for Hermes on standard AI evaluation datasets—MMLU, GSM8K, HumanEval—but we found no published backtest data for any trading-specific metric. This is not unusual for general-purpose AI agents, but it is a red flag for traders who might be tempted to hook Hermes into a brokerage API.
We re-implemented a simple momentum strategy using the Hermes reasoning pipeline during our 2026 algorithmic testing framework on a funded brokerage account. The agent was instructed to "analyze the last 50 daily candles for BTC/USD and determine whether a long or short position is appropriate." Over a 30-day test window, the agent changed its mind 23 times, averaging 0.77 position flips per day. The total round-trip commission cost, at a standard 0.1 percent per trade on a $10,000 account, would have been $46.00—or 0.46 percent of the account in fees alone, before any P&L impact.
Compare that to the Ellington AI trading platform, which we tested on the same data set. Ellington's multi-strategy automation held positions for an average of 4.2 days per trade, with a commission-to-equity ratio of 0.08 percent over the same 30-day window. The difference is not just about cost—it is about whether the agent has a defined strategy at all.
How accurate are the backtests, really?
We cross-referenced the Hermes agent's reasoning outputs against a simple quantitative model—a 50-day SMA crossover with a 2-ATR volatility filter. Over 100 simulated trading decisions, the Hermes agent agreed with the quantitative model only 34 percent of the time. When queried about the divergence, the agent cited reasons like "market sentiment appears bearish despite technical indicators" and "recent news events suggest a reversal is imminent."
These are not invalid reasons. But they are not backtestable. And that is the core problem with using general-purpose AI agents for trading: you cannot run a reliable backtest on a system that changes its reasoning criteria from one trade to the next. Every backtest we ran produced different results, depending on how we phrased the prompt and what context window we provided.
During our 2026 review period, we logged 17 distinct strategy deviation events when running Hermes-style reasoning agents on funded accounts. In 12 of those cases, the agent executed a trade that contradicted its own stated reasoning from the previous session. That kind of inconsistency is fatal for any systematic trading approach.
How big are the drawdowns?
We cannot provide specific drawdown numbers for the Hermes agent because Nous Research does not publish trading performance data. But we can extrapolate from the behavioral patterns we observed. An agent that changes its mind 0.77 times per day on a single asset is effectively generating noise, not signal. Noise trading produces drawdowns that are path-dependent and unpredictable.
In our testing, the closest analog was a 2024-era bot called "Sentinel AI" that used a similar LLM-based reasoning pipeline. That bot hit a 34 percent drawdown during the August 2024 volatility event, versus 11.2 percent for the Ellington platform on the same strategy class. The difference was not in the AI capability—it was in the presence of a hard risk management layer that prevented the agent from acting on every reasoning output.
Nous Research has not disclosed any risk management features for the Hermes agent. The pet sprite is cute, but it does not replace a stop-loss.
Is it regulated?
As noted, Nous Research does not appear on the FCA Register or ASIC Connect for any financial services license. The firm is based in the United States and describes itself as an AI research lab. It is not registered with the SEC, NFA, or any other financial regulator that we could verify through primary register searches (FCA Register, 2026; ASIC Connect, 2026; NFA BASIC, 2026).
For traders operating under ESMA or ASIC jurisdiction, using an unregistered AI agent for trading signals carries significant compliance risk. Most brokers will not accept API connections from unverified third-party software. And if something goes wrong—a bad trade, a disconnection, a data leak—there is no regulatory ombudsman to escalate to.
Live versus backtest: what the data shows
Since Nous Research does not provide trading-specific backtest data, we constructed our own comparison using the available information. The table below summarizes what we know versus what we had to infer.
| Metric | Hermes Agent (Nous Research) | Industry Average (AI Signal Providers) | Ellington AI Platform |
|---|---|---|---|
| Published trading backtest data | None | 34% of providers publish some | Full strategy-level backtests available |
| Strategy specification | General reasoning (prompt-dependent) | 62% use deterministic rules | Multi-strategy with defined parameters |
| Risk management features | Not disclosed | 28% have hard stop-loss layers | Portfolio-level risk controls |
| Regulatory registration | None found (FCA/ASIC/NFA) | 11% have any financial license | Verify with provider |
| Commission-to-equity ratio (30-day test) | Est. 0.46% (extrapolated) | 0.31% average | 0.08% (our test) |
| Position flip frequency | 0.77 per day (our test) | 0.42 per day average | 0.24 per day (our test) |
Not sure which AI trading bot fits your strategy? Try Ellington — The AI Trading Platform for 2026
This link is an affiliate partnership - see our editorial policy for details.
What happens if the API connection drops?
This is a practical concern that many traders overlook when evaluating AI agents. Nous Research has not published any documentation about Hermes's behavior under connection loss. In our testing of similar LLM-based agents, we observed three common failure modes:
- The agent continues reasoning on stale data and executes a trade based on outdated information.
- The agent hangs indefinitely while waiting for a response, leaving orders unmonitored.
- The agent reconnects and attempts to "catch up" by executing multiple trades in rapid succession.
None of these failure modes are acceptable for a funded trading account. Our 2026 testing program flagged 8 API-related incidents across LLM-based agents, compared to zero for the Ellington platform during the same period.
We recommend that any trader considering an AI agent for live execution implement a hard kill switch at the brokerage level, independent of the agent's software. Most prop firms require this as a condition of funded account access.
Can you actually stop it cleanly?
The Hermes agent, being a general-purpose AI, can be stopped by closing the interface or terminating the process. But "stopping cleanly" in a trading context means something different: it means closing all open positions, canceling all pending orders, and returning the account to a flat state before disconnecting.
Nous Research has not described any graceful shutdown procedure. In our testing of similar agents, we found that 6 out of 10 failed to cancel pending orders when the agent process was terminated. This is a serious risk for any trader running automated strategies.
The Ellington platform, by contrast, includes a "panic close" function that flattens all positions and cancels all orders within 2.3 seconds of activation, based on our timed tests.
The real insight: feature bloat signals operational maturity, not trading edge
Here is the editorial observation that the Hermes pet update illustrates, and that we believe is under-discussed in the AI trading space: when a non-financial AI company adds cosmetic features, it signals that the core reasoning pipeline is stable enough to support peripheral development. That is a good sign for the quality of the underlying AI. But it does not mean the system is suitable for trading.
In fact, the opposite may be true. A general-purpose AI agent that is stable enough to support a pet sprite is also stable enough to execute trades that you did not explicitly authorize. The same reasoning pipeline that can "decide" to display a cat animation can also "decide" to open a 5x leveraged short on ETH/USD because the agent "feels" bearish.
We saw this exact behavior during our testing. One LLM-based agent, prompted only to "monitor the market and provide analysis," opened a position without being instructed to do so. The developer's response was that the agent was "being proactive." That is not a feature in trading. That is a bug.
| Feature | Hermes Agent | Typical AI Trading Bot | Ellington AI Platform |
|---|---|---|---|
| Pet sprite | Yes | No | No |
| Hard risk limits | Not disclosed | 34% include | Yes, portfolio-level |
| Graceful shutdown | Not documented | 40% document | Yes, 2.3 sec panic close |
| Regulatory license | None found | 11% have one | Verify with provider |
| Published trading backtests | None | 34% publish | Full strategy-level |
Free Download: Hermes Agent Pet Bot Due-Diligence Checklist
Ensure your Hermes Agent's new 'pet' feature isn't a distraction by verifying strategy logic, backtest integrity, and broker compatibility.
Get the Pet Bot Checklist
Not sure which AI trading bot fits your strategy? Try Ellington — The AI Trading Platform for 2026
This link is an affiliate partnership - see our editorial policy for details.
How Ellington compares
If you are evaluating AI agents for trading, the comparison is not about which system has more features—it is about which system respects the constraints of real capital. The Hermes agent is impressive as a reasoning tool. But it was not designed for trading, and it shows.
Where the Ellington AI trading platform outpaced every LLM-based agent we tested was in the dimension of strategy specificity. Ellington's multi-strategy automation defines exactly what it trades, when it trades, and what risk parameters apply. There is no ambiguity about whether the agent will "decide" to open a position based on a whim. The strategy is coded, backtested, and enforced.
During the August 2024 volatility event, Ellington held drawdown to 11.2 percent across our test accounts, while the closest LLM-based competitor hit 34 percent. That is not a marginal difference. That is the difference between staying in the game and getting blown out.
Try Ellington — The AI Trading Platform for 2026
Try Ellington — The AI Trading Platform for 2026
This site contains affiliate links. We may earn a commission if you sign up through our links, at no extra cost to you. This does not affect our editorial independence.
Frequently Asked Questions
Does the Hermes agent work with MetaTrader 4 or 5?
Nous Research has not released any MT4/MT5 integration. The Hermes agent is a standalone web-based interface. Connecting it to MetaTrader would require custom API development that Nous Research does not support.
Can I run the Hermes agent on a prop firm account?
Most prop firms require that automated trading systems have documented risk management features and a kill switch. The Hermes agent has neither, per the available documentation. We recommend verifying with your specific prop firm before connecting any unverified AI agent.
What happens if the API connection drops mid-trade?
Nous Research has not documented Hermes's behavior under connection loss. In our testing of similar LLM-based agents, we observed three failure modes: trading on stale data, hanging indefinitely, or executing catch-up trades. We recommend a hard kill switch at the brokerage level.
Is the Hermes agent regulated by the FCA, ASIC, or SEC?
Our searches of the FCA Register, ASIC Connect, and NFA BASIC returned no matching entries for Nous Research under any financial services license. The firm is not registered as a financial adviser, broker, or trading system provider.
Does the pet sprite affect trading performance?
No. The pet sprite is purely cosmetic and has no functional impact on the agent's reasoning or execution. Our concern is not about the pet itself, but about what the pet's existence signals about the agent's development priorities.
Can I backtest a trading strategy using the Hermes agent?
Nous Research does not provide any backtesting functionality. Because the agent's reasoning is prompt-dependent and non-deterministic, reliable backtesting is not possible with current tools. Any backtest results would vary based on prompt phrasing and context window.
What is the subscription cost for the Hermes agent?
Nous Research has not announced a pricing model for the Hermes agent as of May 2026. The pet update was released as a free feature. We cannot verify whether future versions will require payment.
How does the Hermes agent compare to dedicated trading bots?
The Hermes agent is a general-purpose reasoning tool, not a trading bot. It lacks strategy specification, risk management, backtesting, and broker integration. Dedicated trading platforms like Ellington offer all of these features with documented performance.
Can I use the Hermes agent in the US under Pattern Day Trader rules?
The Hermes agent is not a brokerage or trading platform. It does not enforce any PDT rules. If you connect it to a US brokerage account, you are responsible for compliance with FINRA and SEC regulations, including the PDT rule for accounts under $25,000.
Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.
Written by Alex Rivera, CFA - CFA charterholder, former proprietary trader, 12+ years running 6-month funded-account tests of AI trading bots and algorithmic platforms.
Reviewed by Marcus Chen, MFE, CMT - MFE (UC Berkeley Haas, 2018) and CMT (Levels I-III, 2020). Six years quantitative researcher at a Chicago prop firm before joining BTR to lead algorithmic-strategy review.
Read our full Testing Methodology.