How to use Hermes Agent to cut trading token costs
The author reports AI calls dropping from 7,200 per day to about 960 and daily token costs changing from $8–$52 to $3–$10. No trading-return improvement is reported.
dev.to
Prompt to try
Editorial adaptation, not the source author’s exact prompt. Replace bracketed inputs and review tool permissions first.
Inspect [Hermes Agent workflow] for unnecessary model calls. Remove unrelated skills, compute [technical indicators] before AI analysis, and route only confirmed signals onward. Apply [scan interval], [call limit], and [token limit], retaining [risk gates] and reporting usage.
Reach people exploring AI agents
See placements →What happened
The author reports AI calls dropping from 7,200 per day to about 960 and daily token costs changing from $8–$52 to $3–$10. No trading-return improvement is reported.
Reported by j_dev28 on dev.toHow the agent did it
- Removed irrelevant skill categories from the agent.
- Built a computational multi-timeframe technical-analysis filter before AI research.
- Reduced scan frequency, calls per cycle, and tokens per call.
Where a person stepped in
The post does not describe human trade approval or review steps; it reports automated filtering and ten risk checks.
A person set the goal and reviewed the result.
The author built a continuously running trading agent that scanned markets, analyzed indicators, and executed trades when signals aligned. Its original architecture invoked expensive AI analysis after triggers even when signals were weak. The supplied post describes an engineering change to reduce those calls, rather than demonstrating better trading returns.
First, the author removed irrelevant skill categories from an agent that had been loading more than a hundred skills. They then added a computational technical-analysis filter before AI research. It checks multiple timeframes using EMA crossovers, RSI, ADX, ATR, volume confirmation, and trend alignment. Only signals classified as confirmed proceed to model analysis.
They also adjusted operating limits: scans moved from sixty seconds to three minutes, the trigger threshold increased, maximum AI calls per cycle dropped, and the token allowance per call was halved. News fetching was removed. The resulting architecture separates heartbeat scanning, statistical triggers, technical filtering, AI research, and ten execution risk checks.
The author reports AI calls falling from 7,200 per day to roughly 960, with daily costs changing from $8–$52 to $3–$10. The stack includes Hermes Agent, Next.js, Hyperliquid, and OpenRouter. The post does not describe human trade approval or code review.
Who it's for: Engineers inspecting repeated AI-call costs in an existing automated system.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: dev.to · j_dev28
Frequently asked questions
What ran before AI research?
A computational filter checked EMA crossovers, RSI, ADX, ATR, volume, and trend alignment across timeframes. Only confirmed signals proceeded to AI analysis.
Did the post report improved trading returns?
No. The supplied text reports reduced AI calls and token costs, along with the revised architecture, but supplies no trading-return comparison.
More details
Remaining steps
- Removed news fetching and retained risk checks before execution.
About this account
The system scans markets continuously and reserves AI research for statistically confirmed signals before execution gates.
Platform: dev.to. Attribution: https://dev.to/j_dev28/i-slashed-my-ai-trading-agent-token-costs-by-80-heres-the-architecture-5292. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: j_dev28 on dev.to
Source published 12 May 2026
Source checked 14 Oct 2026
Tools and topics
Hermes Agent · Next.js 16 · Hyperliquid API · OpenRouter
Token costs · Signal filtering · Hermes Agent
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.