Registered Agent Builders

I want to build a market-making agent that provides liquidity on Hyperliquid perpetuals. My strategy will use dynamic spread adjustment based on volatility and inventory levels...

Alan Coppola

Trading agent description: we have changed this from a narrow-band leverage farmer to a wide-band reaper. The change was based on simulations and real-trading, which showed our original plan was flawed, hence the change. Current Bot Summary: Wide-band Orca SOL/USDC Whirlpool LP on Czfq3xZZ…: 16% width, 0.5 rebalance threshold. Offline grid (R002) shows the old 1% band loses −$557 (no-go); sit-wide 16% finishes +$29 / +$25 on train/holdout (absolute no-loss). Race YAML matches that economics. Volume = fees-implied traded volume. LLM does not trade. Tags: market-making lp orca solana gateway whirlpools clmm simulation Exchanges: Orca (primary). Eligible venue: Solana. Description This agent holds one concentrated SOL/USDC position on Orca Whirlpool Czfq3xZZDmsdGdUyrNLtRhGc47cXcZtLG4crryfu44zE (0.04% fee). Execution is official Hummingbot V2 lp_rebalancer + Gateway orca/clmm. Simulation gate (required): Losing money in simulation is a no-go. Same Gecko hourly path, $800 start, 1× share:

Racing for

Orca

Spenser Wu

student

NYC

I want to build a market making bot for a CEX, likely HYPE, OKX, or LIT. The plan is to use an agent to constantly tune the hyperparameters of the market making strategy we decide on, possibly a modified version of the classic Avellaneda Stoikov model. We want to monitor volatility, basic imbalance, spreads, and other features we research during the hackathon and possibly roll them into one "super-signal" that dictates how we configure our market making strategy.

Racing for

Botcamp

Building an autonomous liquidity agent for tokenized RWAs on Orca. It adapts pricing, liquidity placement, and risk as market conditions change across the 24/7 trading cycle, with the goal of making onchain RWA markets more efficient when traditional market structure breaks down.

Racing for

Orca

Frank Gusto

Platform Engineer

Berlin

I want to transfer my polymarket market maker bot to Derive and check if I can do 5 minutes up down market arb

Racing for

Derive

I'm building a market making bot on Derive for perpetuals. The idea is to create a general purpose mm that is asset agnostic and situation agnostic, taking signals from options to adjust its processes.

Racing for

Derive

Augustya S

"I want to build an autonomous market-making agent for perpetual contracts (e.g., Hyperliquid / Derive) that dynamically adjusts bid/ask spreads based on real-time order book skew, ATR volatility, and inventory risk. The agent will integrate a deterministic hard stop-loss and rebalance inventory through TWAP/VWAP execution when exposure thresholds are breached."

Racing for

XRPL

Shamu II - Liquidity Hunter: An Orca CLMM strategy for RWA pools Shamu stopped chasing price. Now it hunts where price is going. Shamu v1 was born in a Hummingbot botcamp years ago - an algorithm that rapidly shifted LP positions across Orca's concentrated liquidity buckets to keep tight fee-earning brackets around price. It worked, and it earned attention, but it also bled: chasing price means constantly arriving late, and impermanent loss is the toll for that. Shamu has matured. A young orca hunts by outrunning prey in open water. A mature orca studies the migration route and waits at the strait the prey has to pass through. The kill isn't a sprint, it's geometry. Liquidity Hunter applies that logic to Orca's RWA pools. Price isn't noise to chase, it's prey drawn toward specific liquidity levels. When price reaches one, it triggers a raid. It causes dense trading activity and consolidation in a tight range, before price expands toward the next target. v1 stayed wrapped tight around wherever price was. v2 anticipates where price is headed and pre-positions brackets there before the raid begins, holding tight, tick-optimized ranges through consolidation, and stepping out of the water during expansions that offer nothing but IL. Same apex predator, same name. Different hunt. Shamu no longer chases the current, it knows where it's going, and it's already there waiting.

Racing for

Orca

Nazir Adams

I plan to build an agent that identifies patterns within the market, double tops, head n shoulders etc. By putting together a playbook to identify these types of patterns one can easily identify the patterns as they happen. The agent will be able to run backtests on any pair on any exchange and build up data to more accurately identify the patterns for a particular pair and determine what the best average TP/SL/TL is. The agent can then trade the patterns when they appear,

James LO

Autonomous concentrated liquidity market-maker on Meteora Dynamic Liquidity Market Maker (DLMM) on Solana. Each tick, it scans Meteora DLMM pools via GeckoTerminal, selects the pool with the highest volume-to-total-value-locked (V/T) ratio that passes structural and routability checks, rebalances the wallet via Jupiter if needed, and deploys a managed liquidity-providing (LP) position — closing it automatically when defined exit conditions are met.

Racing for

Botcamp

David Pere

solutions engineering schizo

Lagos

Derive CESF Crash-Mass: Hummingbot V2 controller + Condor agent that buys cheap vol when HAR-RV/EWMA forecast > SVI ATM IV and CESF crash-mass says downside is operationally distinguishable. Perps proxy + Black76 options, Kelly + Guard, 1h, 4-pair universe.

Racing for

Derive

Casmir Patterson

Quantative Developer

Chicago

I want to build a market making bot for SAGA->Solana to provide liquidity for my BEZY Token, Using a token buy back and volume bot strategy to accumulate supply and stabilize and support our market cap.

Racing for

Botcamp

Anonymous Builder

I want to build an autonomous, AI-driven spot trading agent designed for consistent capital preservation and smart liquidity provisioning. The agent leverages LLMs via the Condor Agent Harness to dynamically adapt to shifting market conditions while maintaining strict, automated risk boundaries to protect the trading capital under all circumstances.

Racing for

Bitget

Southen_ .

iCode

Remote

### 📝 Clean Strategy Description (Copy & Paste Ready) **Strategy Type & Protocol:** An autonomous, volatility-adaptive Concentrated Liquidity (DLMM) market-maker on Solana via Meteora DLMM pools, connected through Hummingbot Gateway with cross-venue delta-hedging on perpetual markets (Bitget / Gate.io / Hyperliquid). **Core Mechanics:** Rather than deploying static bin ranges, the agent dynamically prices bin spreads in Realized Volatility units (Garman-Klass / Parkinson RV). It switches between symmetric Gaussian Curve distributions during mean-reverting consolidation (maximizing fee capture per dollar) and momentum-skewed BidAsk distributions during directional trending flow. **Cross-Venue Hedging (Delta-Neutral Yield):** As price moves across bins and spot inventory shifts, the controller continuously tracks net portfolio delta in real-time and executes low-latency micro-hedges on perpetuals to maintain 100% delta neutrality. This transforms concentrated LPing into a pure fee-harvesting engine insulated from underlying token drawdowns. **What Makes This Unique:** 1. **Volatility-Engineered Bins:** Dynamic bin range expansion and compression calibrated to expected bin dwell time rather than arbitrary fixed percentages. 2. **Economic Churn Gate:** Prevents the "rebalance whip" by enforcing that Expected Incremental Fees must exceed (Slippage + Solana Priority Fees + Hedge Rebalance Costs), backed by multi-slot dwell time verification. 3. **Fail-Closed Safety Engine:** Built with the Hummingbot V2 Controller architecture, featuring automated markout telemetry, stale RPC circuit breakers, and hard inventory floor stops.

Racing for

Meteora

John Schugart

subject to change: A cross-venue market maker on Derive -PERP, built as a Hummingbot V2 controller with a Condor routine supervising it. Fair value comes from the deepest same-currency leader, basis-adjusted, with volatility from a slow window. Quotes rest in a ladder around that fair value: tight levels for volume, wide levels for edge. Every level is re-quoted when the leader moves, and every fill is hedged or marked back to the leader.

Blas Palmisciano

Mar del Plata, Argentina

An agent run fleet of Hummingbot market making bots, continuously improved by agents that backtest candidate configs at 1 second fidelity on a live identical local engine and deploy the winners. It's unique because it creates an autonomous optimize → deploy → monitor loop

Racing for

Botcamp

Strategy type: A DEX-anchored market-making agent. Instead of pricing off the order book it's quoting into (which for a thin or newly listed token is empty, one-sided, or ours), the agent takes its fair value from the on-chain spot price of the asset's deepest Uniswap V3 pool, read directly from slot0 over RPC, and quotes a symmetric ladder of post-only limit orders around it on a centralized-style order book. In effect it exports on-chain liquidity to venues that don't have any yet. Markets / exchanges: Hyperliquid spot, AsterDEX spot, with Uniswap V3 on Arbitrum as the price oracle. Built for long-tail tokens that are liquid on-chain but have empty CEX/perp-DEX books — we bootstrapped the first two-sided market for a newly listed token from a literally empty book and have run it live since August 19 at up to $300k per side. What makes it unique: 1. The anchor is the DEX, not the venue. Reference quality is graded every tick (ok / degraded / stale / none) by comparing the pool price to whatever third-party liquidity exists on the venue after stripping our own orders and dust; degraded → every quote widens, stale or too far from the venue → all quotes pulled. The agent also measures how much capital it would cost to move the anchor pool 0.1% / 1% and logs it as its own risk, because a manipulable oracle is the real threat to this strategy. 2. Inventory-aware, not just symmetric. Quotes skew in price and size toward a target base/quote value fraction, bounded by drift brackets around actual holdings, with per-side burst-fill pauses to defuse pick-offs and a daily-loss latch that survives restarts. 3. Built for the venue's real constraints. Batched place/modify/cancel, integer-tick price math, deterministic client order IDs, a single serialized nonce queue, exponential backoff on venue refusals, and a triple dead-man (scheduled cancel + SIGTERM cancel-all + an independent watchdog process that cancels everything if the engine's heartbeat goes stale). 4. It's an Agent, not a script. The market maker is one strategy an LLM-powered trading Agent can propose from chat, priced against the user's policy engine and started only through a human "slide to approve." The same engine runs sibling strategies — a DEX↔CEX arbitrage taker and a paced round-trip volume loop — and an orchestrator Agent runs a fleet of them across dozens of exchange accounts as a strategy arena. No LLM is ever in the trading loop; the model decides what to run, deterministic code decides every order.

Racing for

Botcamp

Roman Kurnovskii

Software Engineer

Israel

I want to build a market-making agent that provides liquidity on Hyperliquid perpetuals contracts. This agent will extend Etemaro's core ReAct (Reasoning + Acting) LLM agent framework to operate on Hyperliquid's perpetual swap markets. The agent will continuously monitor market conditions, adjust bid/ask spreads dynamically, and manage inventory risk to provide liquidity while capturing spreads.

Racing for

Meteora

BitBrainy

Bengaluru, India

Meteora DLMM auto-rebalancer on Solana. I will provide liquidity on one liquid Meteora pool (e.g. SOL/USDC or another major pair), not long-tail memes. The agent keeps the LP range around the current price. When price leaves the range, it closes and reopens only if expected fees beat rebalance cost (gas + slippage). Hard stop on drawdown and failed transactions. Built with Condor + Hummingbot LP executor + Gateway Meteora connector. Optional later: hedge leftover inventory on a perp venue.

Racing for

Meteora

An options-aware adaptive grid for Derive perpetuals,

Racing for

Derive

I am building a Condor Agent that orchestrates Hummingbot V2 execution for volatility-scaled market making on Bitget USDT-M perpetuals — not a discretionary LLM trader. The agent quotes both sides with spreads and sizes driven by realised volatility and inventory skew, then measures short-horizon markout, fees, and adverse selection after each fill. It only tightens quotes when observed net edge supports it, and switches between pre-tested tight / normal / defensive / paused modes when conditions deteriorate. Hard capital limits (inventory caps, drawdown halt, no runaway leverage, stop on stale data) are enforced in code independently of the LLM so the bot can run unattended for the 48-hour finals. Condor is already deployed on my VPS; I will submit the Condor agent, strategy.md, and a live demo before the freeze.

Racing for

Bitget

Aven

I want to build a market making agent in XRPL.

Racing for

XRPL

My agent will use a strategy that provides concentrated liquidity on meteora, hedges on hyper liquid, and automagically rebalances utilizing a bridge i have not decided yet.

Jim king

trader,maker

beijing

I want to race for Orca with a Condor Agent that orchestrates Hummingbot execution on Orca Whirlpools — not a discretionary LLM trader. The agent scans eligible Orca CLMM pools, ranks them by fees, volume, volatility and token risk, and only opens a concentrated LP when expected fees justify the risk. Positions are executed through Hummingbot Gateway / LP Executor. The agent re-ranges when price leaves the active ticks, and exits on excessive volatility, drawdown, or pool-quality deterioration. Hard capital limits (no leverage, inventory and drawdown caps, halt on stale RPC) are enforced in code independently of the LLM, so the bot can run unattended for the 48-hour finals. I will submit the Condor agent, strategy.md, and a live demo before the freeze.

Racing for

Orca

An autonomous market-making agent for Meteora DLMM pools, rebalancing concentrated liquidity positions in real time to maximize yield.

Racing for

Meteora

Р04

Kazakhstan

I want to build strategy that plays against market in spot

KOMARI Subheeksh

india

XRPL AMM liquidity agent built on Condor's xrpl_market_maker scaffold. Deploys the $800 into the XRP/RLUSD AMM pool with an automated fee-compounding loop (re-adds accrued fees every N ledgers) and a passive CLOB quote layer that offers inside the AMM curve on both sides. Hard risk controls: halt on –3% drawdown, inventory bounded ±$200, no leverage, no directional signals. Volume is generated by counterparty swaps against the LP position and by CLOB fills, not by taking spread. Goal is high turnover-per-dollar with variance clamped near zero — the same shape as Cohort 13's winning agent, adapted to XRPL's near-zero fee environment where $800 is not thin capital.

Racing for

XRPL

memeshe

Da Nang, Viet Nam

I want to build an CLOB↔AMM dual-venue market maker on XRPL.

Racing for

XRPL

ace

Anambra, Nigeria

I want to build BlackBox an autonomous LP risk engine for Meteora. It doesn't blindly chase fees. It predicts when liquidity becomes the wrong position, simulates alternatives, and moves - or retreats - before the market forces it to.

Racing for

Meteora

Cedar

I am building directional strategy

Piotr Wasiel

Vibe-Quant-Trader

Żywiec, Poland

I want to build a market-neutral spot–perp basis agent on Bitget. It will monitor multiple liquid markets, identify unusually wide basis relative to a rolling anchor, and open delta-neutral long spot / short perp positions when the expected basis capture and funding outweigh fees and execution costs. The agent will dynamically select markets and allocate capital based on net executable edge, while using passive spot orders with immediate perp hedging to control slippage and directional risk.

Racing for

Bitget

Golden

Japan

directional agent

Racing for

Derive

Mirasol

Singapore

Market making agent

roux

Philippines

hedge strategy

Elle

Singapore

market-making agent, details to be revealed in official submission

Racing for

Bitget

ctrader xt

Philippines

Market making both spot and perps

Racing for

Gate

Jadonamite Kenechukwu

Creativity Peaked

Africa , nigeria

A market-making agent on Hyperliquid perpetuals — with the edge in position sizing rather than quoting. Dynamic spread adjustment on volatility and inventory is table stakes; every serious entrant will have it. Most 48-hour races aren't lost on bad spreads, they're lost to inventory blowup. So my differentiator is a capital-pacing controller I've already built and tested, which ports TCP congestion control — Google's BBR — to capital deployment. Instead of trading until it hits a risk limit (the loss-based behaviour of 1980s TCP), it continuously models a ceiling from measured value-rate and result-latency, paces inventory below it, and probes upward only when the model says there's room. Underneath sits a hard floor — per-position, total inventory, rolling-24h — that holds regardless of what the model believes; a test fires 10,000 retries at it and proves it cannot overspend. The transplant has precedent: Netflix took BBR's insight out of the network and into RPC concurrency limits. This is the second hop — requests to capital. Volume comes from quoting both sides continuously; survival comes from the controller. In a race scored on volume and P&L over a fixed window, the agent still standing at hour 47 wins.

Racing for

Botcamp

Nivesh Gajengi

UAE

i want to make Trading Strategy bot that can be used to backtest trades and then create stratergies to use for other agents

Racing for

Meteora

Sergiu O

Solo builder, Chisinau. Market microstructure and liquidation data. 25 services in production.

Chisinau, Moldova

quench is a market maker for Bitget USDT-M perpetuals, built as a Hummingbot V2 controller. It quotes both sides in units of realised volatility, and every exit scales with the quote that filled it, so a fill five volatility units away from mid targets its way back toward mid instead of a fixed take profit. That single detail is what decides whether a wide quote pays for itself. On top of the quoting sits a liquidation fuel map. A collector reads open interest changes off the perpetual tape and projects them into leverage-implied liquidation clusters above and below price, marking a cluster spent once the tape has traded through it. The agent will not sell into unspent short-liquidation fuel above it and will not buy into long-liquidation fuel below it. When the feed goes stale the layer switches itself off and the agent falls back to plain volatility-scaled quoting. It never acts on stale data. I ran it inside Hummingbot's own V2 backtesting engine over fourteen days of one-minute SOL data, with 34 offline tests, and I will tell you what the numbers said rather than what I wanted them to say. Quoting one volatility unit wide loses money at a 2 bp maker fee, because gross edge per round trip is 2.3 bp. Quoting five and ten wide earns 5.7 bp gross, clears the fee, and returns a t-statistic of 2.4 across 53 fills with both halves of the sample positive. The liquidation layer as first written cost 11 percent of net, so I cut the component responsible instead of keeping it for the story. Solo builder in Chisinau, running about 25 services in production on my own box with watchdogs. The agent will still be alive at hour 47.

Racing for

Bitget

Due Diligence by Top Traders

UAE, Dubai

looking for new algos for my portfolio

Racing for

Bitget

kenneth umoekpe

nigeria

Regime-Aware Solana Liquidity Agent Build a Condor agent that autonomously provides concentrated liquidity on Meteora or Orca while protecting its capital from volatile or unsafe pools. The agent would: Scan eligible Solana pools. Rank them using fees, liquidity, volume, volatility, and token-risk signals. Detect whether the market is trending, ranging, or becoming unstable. Open concentrated-liquidity positions only when expected fee income justifies the risk. Dynamically widen, narrow, or reposition its liquidity range. Exit when volatility, drawdown, pool quality, or impermanent-loss risk becomes excessive. Explain and log every decision. Enforce hard capital limits independently of the LLM.

jilt jeeltcraft

creative developer

Modena Italy

the agent operates in the omnity ree network with its own LP and flash loan infrastructure, across evm, ICP and Bitcoin networks, atomically.

Racing for

Orca

Half Doctor

United Kingdom of Great Britain and Northern Ireland (the)

The Derive Volatility Spread Trader is an autonomous quantitative agent that systematically harvests crypto options volatility risk premia on Derive. By combining forecast RV vs. IV edge modelling with real-time Dealer GEX intelligence from Derivatives Monkey, it executes atomic 4-leg RFQ option packages and maintains strict delta neutrality via zero-fee perpetual rebalancing.

Racing for

Derive

DeFi Sentinel Trading Agent — an autonomous market-making agent built on Condor that provides two-sided liquidity on Solana DEXs (Orca Whirlpools) with signal-driven range management. The agent reads on-chain order flow and volatility metrics to dynamically adjust spread width, inventory skew, and rebalancing timing. Unlike static LP bots, our agent uses anomaly detection (Sentinel-style) to avoid realizing impermanent loss during volatility flushes, while maximizing fee capture during stable ranges. Built on our MCP/agent stack (DeFi monitoring + Hummingbot execution layer).

Fikan Ali

Full-stack developer

Nigeria

I want to build an adaptive trading agent for Bitget BTC/USDT and ETH/USDT that dynamically switches between high-volume market making, funding-rate capture, and perp-basis opportunities based on real-time volatility, liquidity, inventory, and cross-market conditions. Its edge is adaptive execution: rather than using static parameters, it continuously adjusts spreads, exposure, and strategy selection to maximize risk-adjusted trading volume while protecting capital.

Racing for

Bitget

Anonymous Builder

build a liquidity monitor curator for megaeth & if possible include other chains w/ similar settlement to provide signals for arbitrage opportunities; identify cryptotokens that behave more like cryptocurrency & cryptocurrencies that behave more like cryptotokens.

Racing for

Botcamp

Soumalya Paul

India,Bangalore Urban

I'm building an autonomous liquidity-management agent for Meteora's DLMM pools. It can be deployed in two forms: as a vault that accepts user deposits and manages multiple pools together, or as a liquidity-management contract that handles any single pool. In both cases the agent actively manages liquidity by dynamically structuring bin allocations around the current price. Depending on market conditions, it shifts between a delta-neutral configuration (symmetric bins plus hedging to minimize directional exposure) and a profit-seeking configuration (skewed, tighter bins to capture more fees during range-bound or trending conditions). It continuously rebalances bins based on volatility, price movement, and fee generation to reduce impermanent loss while maximizing yield for depositors.

Racing for

Meteora

riyan

Indonesia

Darwin Trader is an adaptive autonomous trading agent for crypto perpetual markets. It continuously runs multiple deterministic strategies in a live Shadow Arena, including momentum, mean reversion, liquidity vacuum, funding crowding, and volatility breakout. Darwin can switch between existing strategies, safely adjust strategy parameters within predefined bounds, and generate new bounded strategy variants. Every new or mutated strategy must first prove itself in shadow trading before progressing through Challenger → Probation → Active and receiving live capital. An event-driven AI council evaluates regime changes and strategy performance, while hard risk limits and Hummingbot handle deterministic execution, position management, stop-losses, take-profit, and order lifecycle. The initial target is Gate perpetual markets. What makes Darwin unique is that the LLM never directly places trades. Strategies compete on measurable live-market performance, and only strategies that demonstrate superior risk-adjusted fitness can control real capital.

Racing for

Gate

I am building two entries that share one market-making engine on Bitget USDT perpetuals, and differ in who makes the decisions. Nomad (Hummingbot V2 controller) quotes a two-sided ladder on one market at a time and moves between a short, pre-declared list of markets as conditions change. The rules are frozen before the race and nothing retunes itself at runtime. Before it moves it stops quoting, closes out, and confirms it is flat; if it cannot confirm that, it stays where it is. Nomad Mind (Condor agent) runs the same ladder and reads the same market data, but an LLM agent chooses the market and how to quote it, and gives its reasoning. Its choices are bounded by ranges frozen before the freeze; deterministic code checks every choice, brings anything out of range back in, and records what it changed. The agent never touches the exchange, and if the model is unreachable the execution layer carries on under its own rules. Both run unattended: self-starting, self-validating against the venue before the first order, restart recovery, and hard limits on inventory, drawdown, loss, stale data and order rate, all enforced in code. Nomad has been running on Bitget mainnet with real money since 24 September. Because both are given identical market data, the pair is a fair test of frozen rules against agent judgement. Fuller detail, final code and video go up by the 30 September freeze.

Racing for

Bitget

Kristian Mikula

Hungary

I'm building an automated liquidity-management agent for Meteora's DLMM pools on Solana, using a Hummingbot Controller connected through the Gateway connector. It opens a concentrated-liquidity position centered on the current price, monitors the pool continuously, and automatically closes and re-opens the position whenever price drifts outside the active range so capital keeps earning fees instead of sitting idle out of range. I'm starting with a simple, well-tested bin distribution and conservative rebalancing thresholds to limit churn and impermanent loss, with room to add volatility-based sizing as I iterate. This is my first Solana/DeFi build, so I'm prioritizing something simple and reliable over something exotic.

Racing for

Meteora

I want to try to build an agent that will be inform us during meteora pool like when the vol spike or down their limit and others notification functions agent

Racing for

Meteora

Luna

Malaysia

I want to build market making agent

Nirmalandu Das

I want to build an AI crypto trading agent that thinks like a skilled discretionary trader. It will trade BTC, ETH, and highly liquid altcoins by analyzing market structure, liquidity, volume, momentum, volatility, and multi-timeframe trends. The agent will adapt to changing market conditions, avoid low-quality setups, and dynamically manage position size, stop-loss, and take-profit. Its key advantage is knowing when not to trade. Every trade will have a clear reason, risk level, and invalidation point. The goal is disciplined, explainable, risk-adjusted trading not endless signals or unrealistic win-rate promises.

Crypto T

Building a live multi-engine trading desk on Bitget USDT-M perpetuals with Hummingbot V2. Three engines share one account, one journal, and a custom dashboard. Risk limits are in code, not in the model. Engine 1 — Momentum controller (v37_scalp_multi): Scans 24 Bitget perps (crypto, gold/oil, stock perps) on closed 15m candles. Three signal engines vote (momentum, mean-reversion, volatility). A slot opens only when two agree above 0.66. Up to 3 positions, $10 margin each, per-symbol leverage. Triple-barrier exits: 0.8% SL, 1.6% TP, trailing 1.2/0.8, 3h time stop. Engine 4 — Maker (pmm_simple): Two-sided quotes on BTC, ETH, XRP, DOGE. Inventory capped at $100 notional per pair (~$10 margin at 10×). This is the volume engine. Maker inventory does not consume Engine 1 slots. Engine 2 — Condor LLM agent (v37_risk_manager): Independent 300s loop, own 3 slots, 3% daily drawdown halt. Journals the thesis on every tick. Opens opportunistic trades off the maker names. Risk overlay (flatten E1/E4 when the thesis is gone) is in the design; not claiming it as proven yet. What makes it unique: systematic coverage, maker volume, and an LLM agent on the same book — tagged E1 / E2 / E4 in one journal, watched on one dashboard (live scanner, strategy radar, stack, Condor). Already live on Bitget, not a paper mock. Built with Hummingbot 2.16 V2 controllers + Condor. Custom dark dashboard.

Racing for

Bitget

I want to build a market-making agent that provides liquidity on Hyperliquid perpetuals. My strategy will use dynamic spread adjustment based on volatility and inventory levels...

Federico Cardoso

I want to build a set of simple directional strategies that an agent backtest and deploy based on market conditions on a perpetual exchange.

Racing for

Gate

David Solutions

AI engineer

Portugal

I want to build an automated liquidity provisioning agent for ORCA

Racing for

Orca

Nolan

Search for high volatility trades and capturing the movement on trend

Macro Wang

Hangzhou

I want to build a market making agent that provides liquidity on hyperliquid perpetuals.

Racing for

Gate

Jordan Jones

Springdale, United States of America (the)

An autonomous, LLM-driven liquidity-provisioning agent for Meteora that runs a concentrated-liquidity (CLMM) fee-yield core with disciplined capital rotation. The agent runs a single coordinated LP loop on a ~5-minute tick: - LP Layer — Concentrated liquidity positions on Meteora DLMM pools (with Orca/Raydium rotation). Scans trending pools, ranks by fee yield, dynamically adjusts range width based on volatility, and rotates capital through per-slot take-profit/stop-loss (20%). - LLM reasoning — reads market conditions, fee rates, and pool depth to decide which pool to enter, when to widen/narrow range, and when to exit — allocating capital to the highest-yield activity at any moment. Most hackathon agents pick one static strategy. This agent uses LLM reasoning to adapt its LP behavior to live market conditions — the Condor harness separates the reasoning (LLM decides) from execution (Hummingbot places orders), so the agent never misses a fee window while thinking. Built on GenTech's existing multi-chain infrastructure — x402 payments, ERC-8004 identity, and gasless settlement via Q402 — this agent is designed to scale beyond the hackathon into a fully autonomous DeFi operator.

Racing for

Meteora

俊华 陆

Guangzhou, Guangdong, CHN

ABCMM — a market-making agent for XRP/RLUSD on Gate.io (CEX) and XRPL native DEX, run via Condor (LLM-driven decision layer) over Hummingbot execution. Core differentiator: quotes are anchored to Flare's FTSO v2 fair-value oracle (decentralized on-chain price), not the CEX mid that most other builders will reference. This means our quotes are honest relative to on-chain truth, not reflexive to the same feed. Strategy: dynamic spread = f(FTSO mid vs CEX mid divergence, inventory skew toward 50/50 XRP/RLUSD, ATR volatility). Order sizes scale with inventory distance from target. Risk guards: max absolute inventory, max order size, kill-switch if FTSO staleness > 30s, hard daily PnL stop. Plan: ship Gate first (easier infra, proven CEX connector), then mirror the same strategy to XRPL DEX — same code path, two venues. Why I can build it: 2+ years Web3 / smart-contract engineering. Recently shipped a CC-enclave rebalancer on Coston2 (Flare testnet) and a Circle Agent Stack–powered Aave keeper — both with on-chain attested execution. Same primitive pattern, now applied to live trading.

noboru noboru

Event-Aware AI Trading Agent | Liquidity + Momentum + Risk Control

Japan

I am building a volatility-adaptive multi-asset market making controller for BTC, ETH, SOL, and selected altcoins on sponsored exchanges The strategy uses EMA50/200 trend filters, ATR-based volatility bands, RSI divergence filters, inventory skew control, dynamic position sizing, and strict stop-loss rules Risk per trade is capped at 1.5%, with a maximum of 2 concurrent positions and a global drawdown stop

Yezir Hasan

Lagos Nigeria

An agent capable enough to win on any tracks mostly I am targeting the robin hood chain

Ilpo Vaatainen

Hybrid Quantitative Trader

Helsinki

Proprietary hybrid system combining algorithmic generation with AI validation. Proven backtested approach. Risk-first architecture.

Racing for

XRPL

Berg 1ce

mm learner for this compete thx！thx！thx！thx！thx！thx！thx！thx！thx！thx！thx！thx！thx！

Aditya Dargan

New York

Most Orca LPs lose money to bad range management — they sit in static ranges that drift out of zone, or they rebalance reactively right at the worst price, locking in losses at every flush. We read orderflow from Binance to do two things: anticipate where SOL is heading and distinguish a real trend from a capitulation flush. The flow signal drives where we center our range, how tight we make our ticks, and crucially when not to rebalance — holding through exhaustion moves that reactive strategies bleed into. Backtest shows X% more fees and Y% less time out of range vs. reactive rebalancing, with the biggest outperformance during volatile flushes. The framework generalizes to any volatile Orca pool, including RWA pairs Our strategy doesn't eliminate impermanent loss — every CL LP carries IL by design. What it does is (a) capture more fees by staying in range longer, and (b) avoid realizing IL at the worst possible prices by not rebalancing into flushes. Net P&L = fees − realized IL − gas. We win on the fees side and we win on the timing of the IL realization —-- Submission for registration for AgentCup: Project Vision: Intelligent Orca Liquidity Provision via Condor LP Executors My team is building an advanced liquidity provision (LP) agent specifically for the Orca decentralized exchange, utilizing the Condor framework to automate and optimize range management. The Core Problem Most Orca liquidity providers fail due to flawed range management strategies. They either deploy static ranges that quickly drift out of zone as market conditions evolve, or they employ reactive rebalancing logic. This reactive approach is particularly destructive: it often forces rebalances during market flushes, locking in realized impermanent loss (IL) at the exact worst possible price points. Our Solution: Signal-Driven Execution Our agent addresses these inefficiencies by integrating external orderflow data—specifically from Binance—to gain predictive insight into market movement. We use this data to perform two critical analytical functions: Trend Anticipation: Predicting SOL directional bias to center our liquidity range more accurately. Volatility Filtering: Distinguishing between genuine trend shifts and temporary, high-volatility capitulation flushes. This signal-driven approach directly informs the agent’s execution logic. It dictates where to center our range, how to calibrate tick widths, and—most importantly—when not to rebalance. By holding positions through short-term exhaustion moves rather than panic-selling or rebalancing into volatility, we avoid the 'bleeding' effect common to reactive strategies. Implementation with Condor We will leverage the Condor LP Executor framework to handle the lifecycle management of these positions. The Condor executors allow us to programmatically wrap our logic into dynamic management containers. This offloads the heavy lifting of position maintenance to the executor, ensuring the agent remains responsive to real-time signals while maintaining strict control over our LP architecture. The Performance Thesis Our objective is not to eliminate impermanent loss, as CL LP inherently carries IL by design. Instead, we optimize the Net P&L equation: (Fees - Realized IL - Gas Costs). Fee Capture: We stay in range longer by centering our liquidity based on orderflow rather than historical averages. IL Mitigation: We avoid realizing IL at suboptimal price points by deferring rebalances during volatility spikes. Backtesting demonstrates significant improvements in fee generation and uptime within our range compared to standard reactive models, with the greatest outperformance occurring during periods of high market volatility. While developed for SOL-based pairs, the framework is designed to be generalized across any volatile Orca pool, including RWA pairs.

Racing for

Orca

Leo M.

I want to build an adaptive market making agent. The bot will provide two sided liquidity near the touch on liquid perp markets while dynamically controlling spread width, order size, and inventory skew based on real time volatility, directional pressure, and current position exposure. The strategy is a dynamic market maker rather than a static quoting bot. In calm conditions, it tightens spreads and increases participation to maximize fill rate and trading volume. In unstable or one sided conditions, it widens quotes, reduces size, and shifts into defense mode to avoid getting run over by adverse inventory. The agent also includes modest flow aware skewing, allowing it to lean with short term market pressure when conditions are favorable instead of blindly fading every move. The goal is to build a bot that stays active, protects capital, and earns both spread capture and leaderboard relevance over the full race window.

Racing for

Gate

Tomás Gaudino

Mar del Plata, Argentina

A market maker for Orca's concentrated liquidity pools (Whirlpools, Solana) that quotes tight tick ranges right around the mid price — maximizing fee capture per unit of capital — with a portfolio-level inventory policy that governs rebalancing so the strategy never ends up fully long in a sell-off or fully short in a rally.

Racing for

Orca

Kunal Ranjan

MM

India

A funding-aware perpetual market-making agent for Gate.io, built as a Condor agent on Hummingbot. It provides high-volume liquidity on top BTC/ETH/SOL-USDT perps, with quoting that adapts to real-time market and funding conditions rather than price risk alone. Disciplined inventory and risk controls keep exposure bounded through volatile and high-funding regimes, with the goal of robust risk-adjusted returns over a fully autonomous run. Built on a market-making engine already hardened through extensive live multi-pair testing.

Racing for

Gate

Mohammed Zaid

I want to build an autonomous trading agent centered around market regime detection and adaptive decision-making. The agent continuously monitors BTC and broader market conditions to identify shifts between trending, ranging, high-volatility, and reversal environments. Based on the detected regime, it dynamically adjusts its trading behavior, risk parameters, and execution logic instead of relying on a single static strategy. The core edge comes from combining regime detection with rebound and reversal identification. The agent is designed to detect exhaustion moves, oversold conditions, and rapid sentiment shifts, allowing it to capture rebounds and short-term opportunities with predefined take-profit and stop-loss levels. It prioritizes a high volume of trades and fast execution while maintaining disciplined risk management. By combining adaptive learning, regime-aware trading, and rebound-capture mechanisms, the agent can remain effective across changing market conditions without being locked into a single strategy.

Racing for

Gate

Kingsley Ojilere

Lagos, Nigeria

Accountable AI trader with on-chain proof of every decision I will build it for gate and bybit exchange AI trader that proves every decision on-chain

Kevin Chon

I want to build an advanced liquidity provision (LP) agent specifically for the Orca decentralized exchange, utilizing the Condor framework to automate and optimize range management. The Core Problem: Most Orca liquidity providers fail due to flawed range management strategies. They either deploy static ranges that quickly drift out of zone as market conditions evolve, or they employ reactive rebalancing logic. This reactive approach is particularly destructive: it often forces rebalances during market flushes, locking in realized impermanent loss (IL) at the exact worst possible price points. Our Solution: Signal-Driven Execution: Our agent addresses these inefficiencies by integrating external orderflow data—specifically from Binance—to gain predictive insight into market movement. We use this data to perform two critical analytical functions: Trend Anticipation: Predicting SOL directional bias to center our liquidity range more accurately. Volatility Filtering: Distinguishing between genuine trend shifts and temporary, high-volatility capitulation flushes. This signal-driven approach directly informs the agent’s execution logic. It dictates where to center our range, how to calibrate tick widths, and—most importantly—when not to rebalance. By holding positions through short-term exhaustion moves rather than panic-selling or rebalancing into volatility, we avoid the 'bleeding' effect common to reactive strategies. Implementation with Condor: We will leverage the Condor LP Executor framework to handle the lifecycle management of these positions. The Condor executors allow us to programmatically wrap our logic into dynamic management containers. This offloads the heavy lifting of position maintenance to the executor, ensuring the agent remains responsive to real-time signals while maintaining strict control over our LP architecture. The Performance Thesis: Our objective is not to eliminate impermanent loss, as CL LP inherently carries IL by design. Instead, we optimize the Net P&L equation: (Fees - Realized IL - Gas Costs). Fee Capture: We stay in range longer by centering our liquidity based on orderflow rather than historical averages. IL Mitigation: We avoid realizing IL at suboptimal price points by deferring rebalances during volatility spikes. Backtesting demonstrates significant improvements in fee generation and uptime within our range compared to standard reactive models, with the greatest outperformance occurring during periods of high market volatility. While developed for SOL-based pairs, the framework is designed to be generalized across any volatile Orca pool, including RWA pairs.

Tatiana Astahova

indonesia

Safe Yield Agent A conservative trading strategy focused on preserving capital and generating stable returns through disciplined risk management, low leverage, and trading only high-probability market opportunities.

Dmitry Belaventsev

A funding-aware inventory market-making agent for Hyperliquid perp, built on Condor's agent framework. Instead of one static PMM, it runs a fleet of PMM controllers across the most liquid perp pairs and reallocates capital toward whichever pair is paying the most realized PnL per unit of volume, reading Condor's 5-minute snapshots and get_custom_info to detect regime shifts and throttle exposure when a market turns trending. The edge is the funding leg: it biases inventory toward the side funding pays it to hold, earning spread and funding together while staying near delta-neutral.

David Salas

What type of strategy will your agent use? What markets or exchanges will it trade on? What makes your approach unique? I want to build a market-making agent

We want to explore Covered call strategies on options on Derive

I'm building a delta-neutral trading agent on Derive perpetuals that combines funding rate capture with options-informed positioning. The agent dynamically adjusts spread width and inventory limits based on real-time implied volatility from Derive's options markets, using the derive_perpetual connector. Key features: - Multi-collateral margin management across ETH, BTC, and USDC to maximize capital efficiency - Portfolio margin optimization: cross-position netting to reduce margin requirements and increase deployed capital - Options data integration: reads IV surface and skew to anticipate directional pressure before it hits perps - Adaptive market-making: widens spreads during vol spikes, tightens during low-vol regimes - Risk controls: max drawdown limits, position size caps, and automatic deleveraging What makes it unique: most perp market-makers ignore options signals. By incorporating Derive's native options data into a perps strategy, the agent can front-run volatility regime changes instead of reacting to them. The multi-collateral approach lets it hold positions in the assets it trades, reducing unnecessary conversions and improving capital efficiency.

harvest vrp by selling iron condors. this way it has some defined risk approach to it, while earning yield.

awais raza

I want to build a simple trading agent so I can learn how automated trading works. My goal is to understand how a bot reads market data, follows basic rules, and makes trading decisions. I am mainly interested in learning step by step, starting with a basic strategy before adding anything advanced

I want to build a multi-factor order flow trading agent for BTC perpetual futures that combines Open Interest, Volume Delta, Liquidations, and Order Book Imbalance data into high-conviction Long and Short signals. The strategy works by scoring multiple market conditions simultaneously instead of relying on price action alone. Long signals are generated when Open Interest is increasing, aggressive buy-side Volume Delta is positive, short liquidations are accelerating, and the order book shows bullish imbalance with stronger bid-side liquidity. Short signals use the inverse conditions. The agent will use configurable weighting and threshold-based scoring so trades only execute when multiple institutional-flow signals align together. It will also integrate higher timeframe market structure and VWAP filters to avoid low-quality setups and reduce noise during sideways conditions. The system is designed for crypto perpetual futures markets, initially focused on BTC and ETH perpetuals on major derivatives exchanges. My goal is to build an adaptive, data-driven trading agent that detects real leverage-driven momentum and liquidity shifts in real time, while using strict risk management, dynamic position sizing, and automated execution through Condor.

Tonny Lopez

Algorithmic Trader & Microstructure Builder

I want to build a microstructure-driven trading agent for crypto perpetual markets. The agent will analyze order book data, liquidity zones, trade flow, imbalance, and short-term volatility to detect absorption, liquidity sweeps, and execution opportunities. The system combines high-performance data processing in Rust with a Python decision layer. Rust transforms raw market data into structured signals, while Python evaluates those signals to decide whether to enter, avoid trading, reduce exposure, or wait for better conditions. Within Condor, I want to adapt this into an autonomous agent that observes market conditions, generates microstructure signals, applies strict risk controls, and is tested through simulation or backtesting before live deployment.

Israel Ajayi

market Flow

FlowEdge Regime Adaptive Directional Trading Agent FlowEdge is a directional trading agent built on Hummingbot's V2 framework that adapts its behavior based on live market conditions. It trades crypto perpetual futures — primarily BTC-USDT, ETH-USDT, and SOL-USDT on exchanges like Binance Perpetual, Bybit Perpetual, and Hyperliquid. What it does: The agent uses two timeframes simultaneously. Fast 3-minute candles generate trading signals using Candle Flow Imbalance and VWAP deviation. Slow 15-minute candles classify the market regime using ADX into three states: ranging, trending, or extreme. Entries only fire when at least one timeframe confirms a trending regime otherwise the agent sits out entirely. When it does trade, it places three DCA maker limit orders at price levels that scale dynamically with NATR volatility. Calm markets get tight entries, volatile markets get wide entries. Stop-loss and take-profit scale the same way. What makes it unique: The agent has an embedded OODA loop — it tracks its own last 20 trades in a rolling window and adjusts its signal threshold automatically. If it starts losing, it tightens its entry criteria. If it's winning consistently, it loosens back. This self-adaptation runs every tick inside the controller with zero external dependencies no separate LLM process, no external API calls, no Redis or Kafka. It also reads live funding rates on perpetual pairs and applies a directional bias when positioning is crowded, and uses a gradual RSI dampener instead of a binary filter to preserve partial conviction on strong signals. The entire agent is a single self-contained Python file that inherits from DirectionalTradingControllerBase and uses DCAExecutorConfig with MAKER mode — the same proven pattern as dman_v3. No infrastructure setup needed beyond Hummingbot itself. Vision for the Builders Cup: For the hackathon, I plan to wrap FlowEdge Pro as a full Condor Trading Agent with an LLM-powered reasoning layer that can narrate regime changes, send Telegram alerts on state transitions, and accept natural-language parameter tuning commands. The execution layer is already production-ready the Condor wrapper adds the agentic intelligence on top.

Victor Adeleke

Market master

I'll build a trading agent that combines quantitative analysis, real-time market intelligence, and adaptive risk management to trade crypto, The agent will operate on Binance and Bybit. The strategy is a hybrid multi-factor system that combines: Trend-following models to capture medium- and long-term momentum, Mean reversion algorithms for short-term inefficiencies, The agent will analyze multiple data streams simultaneously, including price action, volatility, order-book imbalance, macroeconomic events, and sentiment signals. It will dynamically switch strategies depending on whether markets are trending, ranging, or highly volatile. What makes this approach unique is the integration of: Risk-first architecture — capital preservation is built into every trade through dynamic stop-losses, portfolio exposure controls, and volatility-adjusted sizing. Cross-market intelligence — the system identifies correlations and arbitrage opportunities between crypto markets in real time. Explainable trading signals — every trade recommendation includes a human-readable explanation of why the position was entered, improving transparency and trust.

I want to build a liquidation sniper bot on Hyperliquid and Binance

IBRAHIM ABDULKARIM

I want to build trading agent that just wins money

I want to build a RSI based strategy where i will have a set of taken which i will going to watch and i will trade (short/long) when it reach 60-40 levels Strategy is very simple. Long when RSI close above 60 being over sold means coming out of 40 levels and for short exactly opposite. take short when it coming out of 60 and close below 40 on a given timeframe. for confirmation i am taking a next bigger time frame like if main time frame is 15m then i am taking 1h form confirmation so if its a long call then i check on confirmation time frame is it above 50 on snapshot not waiting for the candle close if short call then below 50. for Exit if its a long call i put the SL at the previous candle low and for Short Exit previous candle high

Kaira Zambo

I want to build a market-making agent that provides liquidity on XRPL via XRPliquid. My strategy is called Delta Raptor which is an autonomous AI market maker that tracked the volume acceleration of 6 pairs on hourly basis thru a routine. The Agent will inspect the report of routine and then provide liquidity on the top 2 pairs that have the highest volume gained at last hour. Delta Raptor will also have a risk management feature called price band which will not allow order placement if the price suddenly drops or exceeds 2% from starting price. It will have an Auto Rebalancing feature that will trigger whenever an asset has 60% or more. To minimize LLM cost, Deepseek is implemented thru PydanticAI.

Racing for

XRPL

ac wq

我将要构建的交易代理命名为 **“PerpStrat-X”** ，一个专为加密货币**永续合约**设计的多策略自适应交易系统。它的核心设计理念不是寻找圣杯般的单策略，而是让多种低相关性的子策略动态配合，同时在**资金费率、市场微观结构和链上情绪**等永续合约特有维度上建立优势。 --- ### 1. 交易市场与交易所选择 代理将部署在以下市场，兼顾流动性、去中心化选项和低延迟： - **中心化交易所（主战场）** Binance Futures、Bybit USDT Perpetual、OKX Perpetual Swap 选择理由：USDT或USDC本位永续合约流动性最好，交易对全面，API稳定，支持 WebSocket 实时行情和多种高级订单（止损限价、冰山委托等）。 - **去中心化永续协议（辅助套利与备选）** Hyperliquid、dYdX v4、GMX（V2） 理由：链上永续合约提供了不同于 CEX 的流动性池定价，经常出现与 CEX 的价格偏离，这构成独特的套利窗口。同时，交易记录完全链上，有利于策略透明化回测。 代理会同时维护多个交易所的账户和仓位，并通过统一的内部行情总线（price bus）对跨所价差、资金费率差异、深度不平衡做实时监控。 --- ### 2. 核心策略矩阵（四引擎结构） 整个代理由四个相互独立的子策略引擎构成，顶层有一个动态资本分配器决定各引擎的资金权重。 #### ① 自适应趋势追踪引擎（Adaptive Trend） - **逻辑**：使用 Donchian 通道突破 + 波动率调整移动平均（VIDYA），捕捉 1h~4h 级别趋势。 - **永续合约特化**：利用**资金费率乘数**过滤信号。当趋势方向与资金费率方向一致时，增加头寸（说明趋势有真实买盘支撑）；当价格新高但费率极端负值（空头拥挤）时，则只会轻仓跟随，避免轧空回调伤害。 - **退出机制**：结合跟踪止损和波动率目标仓位，每日根据 ATR 调整合约张数，恒定风险预算。 #### ② 资金费率回归/爆发引擎（Funding Rate Mean-Reversion & Trap） - 监控所有交易对永续合约的 8 小时资金费率 Z-score。 - **均值回归模式**：当资金费率处于历史极值（比如高于 +0.1% 或低于 -0.1%），且价格与费率背离（费率极高但价格滞涨，费率极负但价格止跌），发出反向信号，做市式入场赚取费率回归正常和价格反弹的双重利润。 - **资金费率陷阱规避**：如果资金费率极高，但持仓量仍在飙升、多空比继续上升，模型会判定为“资金费率陷阱”——此时不做反转，甚至配合趋势引擎加仓。这是避免盲目套费率爆仓的关键。 - **费率爆发模式**：当新上币或事件导致费率在短时间内巨幅波动，代理会利用期权式思维，做多波动率。例如同时在两个方向上部署突破挂单，赚取价格在费率极端化后的剧烈运动。 #### ③ 统计套利 & 板块配对引擎（Statistical Arb Pairs） - 针对高度相关资产（BTC/ETH、SOL/AVAX、L2 代币对等），使用卡尔曼滤波动态估计对冲比率，构建平稳的价差序列。 - 入场：价差超过 2 个标准差且资金费率差异不会对冲掉预期利润时，做多相对低估永续合约、做空高估合约，保持严格市场中性。 - 独特之处：配对组合会实时计算**跨资金费率成本**。例如做多低资金费率合约，做空高资金费率合约，若持仓时间预期较长，资金费率差可能构成稳定 alpha，而非成本，模型会主动选择这种“顺费率”配对方向。 #### ④ 链上事件 & 订单流驱动引擎（On-chain & Flow Alpha） - 监控链上大额转账（巨鲸运动）、交易所钱包余额变化、流动性池的突然增/减。 - 同时，利用交易所 WebSocket 深度快照，计算**订单簿不平衡指数（OFI）**和**毒性流指标（VPIN）**。 - 当检测到某永续合约突然出现强烈的买方或卖方不平衡，并且链上有对应的大额稳定币/代币转移时，发起动量狙击交易，持仓时间在分钟级，追求捕捉信息扩散前的瞬时价差。 - 此引擎特别适用于 Hyperliquid 等链上协议，其订单簿透明，可直接分析地址行为。 --- ### 3. 方法论独特之处（三大差异点） #### ▍差异化一：资金费率态势感知与动态资本分配 绝大多数代理要么忽视资金费率，要么将其作为独立套利信号。PerpStrat-X 构建了一个 **Funding Regime Classifier（资金费率体制分类器）**，把市场分为四种状态： - 趋势顺费率 - 趋势逆费率 - 费率陷阱 - 费率回归 顶层动态分配器（Bayesian 权重模型或基于近期表现的滑窗夏普最优化）会根据当前体制，调整四个子引擎的风险预算。例如，趋势逆费率阶段，趋势引擎降权，反转引擎提权。这种上下文感知能力极大降低传统策略在极端费率环境下的回撤。 #### ▍差异化二：跨 CeFi-DeFi 实时价值捕获 代理不仅交易单交易所，而是作为一个跨市场参与者，持续扫描 Binance 与 Hyperliquid 之间的永续合约价差。当价差覆盖滑点和提币/操作成本后仍有利润，它会同时在两边建立相反头寸，等价差收敛时平仓，或通过资金费率差长期持仓套取费率时间价值。这是纯粹的 delta 中性策略，为整体组合提供非方向性收益。 #### ▍差异化三：分层强化学习执行与微观结构感知 在下单层面，不使用简单的市价/限价，而是嵌入了一个轻量级 **Soft Actor-Critic 执行智能体**。它在每个时刻根据当前订单簿、价差、近期成交率，动态选择挂单激进程度、是否拆单、是否伪装成冰山订单。训练目标是最小化交易侵蚀 alpha 的冲击成本。执行层还包含“毒性回避”——当市场微观结构出现高频做市商撤退迹象时，代理会主动暂停交易，等待流动性恢复，这在永续合约的插针行情中能救命。 --- ### 4. 全流程风控框架 每个子策略都有以下硬约束，并在代理总控层面汇总： - **最大总杠杆**：动态，根据当前组合波动率和相关性自动计算，通常不超过 3 倍名义杠杆。 - **单币种风险上限**：名义敞口不超过总权益的 20%。 - **策略熔断**：单日、单周亏损达到阈值，对应子引擎自动降权或暂停。 - **资金费率监控**：若总仓位需支付的 8 小时资金费超过预期每日收益的 30%，强制部分减仓。 - **交易所风险分散**：永不将所有保证金存放于单一交易所或协议，使用 API 只读权限与提币限制。 --- 总而言之，PerpStrat-X 不是一个寻找神奇指标的代理，而是一个能理解**永续合约资金费率内部逻辑**、在**中心化和去中心化市场之间架起桥梁**，并通过**微观执行智能体**保护利润的综合交易系统。其最终目标是，在各种市场结构中都能实现稳健、低回撤的风险调整后收益。

I want to build a trading agent that exploits price drift risk on Hyperliquid derivatives during pre-market trading when price deviates from fair value during market close periods.

Minjae Lee

वयधम्मा सङ्खारा अप्पमादेन सम्पादेथ वयधम्मा सङ्खारा अप्पमादेन सम्पादेथ

Ziru Niu

I want to build an order flow-driven market making agent on crypto perpetual futures. The strategy uses dynamic spread adjustment based on short-term volatility (ATR), inventory skewing via CVD and DOM imbalance signals to manage directional exposure, and a hard inventory limit with batch hedging to control drawdown. The goal is a smooth, low-drawdown equity curve through passive liquidity provision rather than directional speculation.

Akeba Clinton

Futures trader

I am a complete beginner in algorithmic trading and I want to build my first market-making trading agent using Hummingbot. I plan to focus on providing liquidity on Dex and Cex perpetuals. My strategy will start simple with basic bid-ask spread management, then gradually add dynamic spread adjustments based on market volatility and my current inventory levels to control risk. I want to learn how to properly manage inventory, avoid big losses, and earn from the spread while trading on a fast decentralized perpetuals exchange. I’m excited to join Botcamp to learn from the instructors and improve.

hula hoops

perpetual_noob

i do not know anything and hope to learn in this process.

Николай Тараданов

1. How to deploy 'deploy'. Right now, even following the instructions doesn't work. 2. Understand the principles of orchestrating multiple bots. 3. Get a backtesting tool. 4. Write an algorithm for removing liquidity from a range.

I would like to build a binary trading agent that learns and constantly improves how to predict the probability of a buy or sell. This will be used to achieve a certain target profit, and win-rate on a daily or regular time scale.

Steven Hudspeth

Cross-venue XRP market maker. Primary venue: XRP/RLUSD on the XRPL native DEX, using Hummingbot order-book strategies with adaptive spread tuned by FTSO v2 fair-value reference. Inventory managed in FXRP and stables on Flare for hedging and yield. Built on top of FlareForward's deployed Apex trading platform on Flare Mainnet. Risk discipline: probe-mode sizing graduates to full bankroll only after live calibration metrics clear. Optional Hyperliquid perp hedge for directional risk control. Stretch goal: XRPL-native control via Flare Smart Accounts so XRPL holders can fund and operate the agent entirely from XRPL.

TANMAY SAYARE

DEVELOPER

I don't have it right now, but I will create a new one and build it .

violain Ot

I want to build a multi-exchange perpetual contract fee arbitrage strategy, including CEX and DEX,

I would like to build various types of ai agents. One example is one which can find gaps between Binance(XRP/RLUSD) and XRPL(XRP/RLUSD) pairs and trade the gaps. One that can measure order flow analysis on the XRPL accurately to take advantage of gaps.

Michael Feng

Botcamp Instructor

San Francisco, CA

I'll productionize the Solana memecoin LP scalping strategy I've shown in podcast episodes.

Racing for

Botcamp

Spencer Ng

An optimized borrowing strategy for creating liquidity events for Derive traders. This strategy leverages Derive's unique and underused favorable lending rate. It pairs it with a long-term short call and a long put on a portfolio to hedge against downside while still unlocking liquidity for personal or investment uses.

Racing for

Derive