Strategy Engine
14 strategies across four buckets — market making, arbitrage, signal/directional, and infrastructure/risk. The quoting engine blends four signals into a single fair value, then sizes around inventory skew.
Hypermes
Welcome to Hypermes.
We build the autonomous trading agent for Hyperliquid perps — 14 strategies, the APEX multi-slot orchestrator, and a nightly REFLECT self-review loop.Autonomous trading agent for Hyperliquid perps — 14 strategies and the APEX orchestrator.
About us
Built around autonomous Hyperliquid trading.
Hyperliquid perps run 24/7 and never sleep. Hypermes ships the agent that quotes, hedges, and reviews itself between ticks — Radar, Pulse, Guard, and a nightly REFLECT loop running without you babysitting it.
APEX Orchestrator
APEX runs two to three live slots in parallel, composing Radar, Pulse, and Guard into one loop. 60-second tick, seven-level entry priority, presets from conservative 2-slot to aggressive 3-slot at 15x.
Hermes Reasoning
Trade decisions backed by Hermes from Nous Research. Market snapshot in, structured order out — every fill carries a reasoning trace.
Architecture
Three layers that make autonomous trading work.
Hypermes is the agent. Hermes is the reasoning. Hyperliquid is the venue. Each layer owns what it’s actually good at.
- Hypermes — the company building software for Hyperliquid perp markets.
- Hermes — the reasoning core, run by Nous Research.
- Hyperliquid — the on-chain perp DEX where every fill lands.
Inside the agent
Six modules.
One autonomous loop.
Each module owns one job. Together they make a single autonomous agent that quotes, sizes, exits, and learns — without a human in the loop.
Quoting Engine
Composite four-signal fair value, dynamic spreads (fee + vol + toxicity + event), inventory skew, exponential-decay multi-level ladder. Powers engine_mm, regime_mm, grid_mm, and liquidation_mm.
Radar
Four-stage funnel scoring every Hyperliquid perp 0–400. Reads market structure, technicals, funding, and BTC macro to surface the few setups worth a slot.
Pulse
Detects sudden capital inflow via OI delta, volume surge, and funding flips. Fires IMMEDIATE signals when the tape genuinely shifts.
REFLECT
Nightly self-review. Computes Win Rate, FDR, direction split, and holding-period buckets, then auto-adjusts APEX parameters within guardrail bounds. The agent gets sharper while you sleep.
Guard
Two-phase trailing stop with tiered profit floors. Posts an exchange-level stop on Hyperliquid, synced on entry and every tier ratchet — survives runner crashes and network drops.
Agent Skills
Ships as a CLI, an MCP server with 16 tools, or a Claude Code / OpenClaw / ClawHub skill — same engine, three interfaces. Built on the open agentskills.io standard.
Research, articles, and
market notes.
Our research
Research desk
Turning raw market data into autonomous trades.
Hypermes studies how prints arrive, how funding rotates, and how strategies decay — then feeds each finding back through REFLECT into the live APEX parameters.
SignalOI, funding, printsSizingInventory + risk
ReasoningHermes by NousGuardExchange-side stop
Why Hyperliquid perps need autonomous reasoning
A breakdown of why human traders still leave money on 24/7 perp markets, and how Hermes reasoning closes the gap on execution and risk.
From single strategy to APEX multi-slot orchestration
A walkthrough of how APEX composes Radar screening, Pulse momentum signals, and Guard exits into 2–3 concurrent slots with a tier-based entry priority.
Why nightly self-reflection beats static parameters
A look at why APEX parameters drift within hours of live trading, and how REFLECT’s nightly review auto-tunes radar threshold, pulse confidence, and daily loss limits against the live tape.
Our case study
Field notes from autonomous trading.
- [01]Wiring Hermes reasoning into the quoting engineHow a Nous Research Hermes call slots into the tick loop without blowing the latency budget, and what the structured output schema actually looks like.
- [02]Designing APEX for multi-slot autonomyHow a single orchestrator manages two to three concurrent positions across uncorrelated perps without stepping on its own inventory or funding exposure.
- [03]From manual MM to fully unattended tradingWhat changes when you stop babysitting the quoting bot every evening and let REFLECT auto-tune APEX thresholds while slot cooldown gates re-entry.
- [04]Reading real liquidity on HyperliquidHow OI delta, depth, and funding shift tick by tick across HL perps, and why composite signals beat any single indicator for fair-value pricing.
- [05]How REFLECT becomes a self-improving loopHow nightly trade-log review turns into a closed feedback loop that re-tunes APEX in production, without overfitting on the last 24 hours.
How we work
How Hypermes ships.
From first signal to live slot — how Hypermes moves from idea to a strategy running on Hyperliquid mainnet.
Research
Study Hyperliquid microstructure, funding cycles, OI rotation, and where Hermes reasoning adds real edge.
Build
Translate findings into one of 14 strategies, the quoting_engine pipeline, and APEX multi-slot orchestration, locked down by a 483-test pytest suite.
Backtest
Replay APEX config variants against historical trade data with scripts/backtest_apex.py, then let REFLECT propose the parameter sweeps that earn promotion to live.
Deploy
Ship to Hyperliquid mainnet via the hl CLI, FastMCP (16 tools), or one-click Railway template. APEX runs a 60s tick loop, Guard syncs an exchange-level stop on entry.
Trading Hyperliquid perps, autonomously.
For early access, partnership inquiries, or technical questions about running Hypermes on your own keys — reach out directly.