Space on š• with Professor Satoshi featuring Pear Protocol & Vooi tomorrow @ 2 PM UTC Set a reminder →

Lit Hub
← Ecosystem

Lobster Agent Trader

Turn the AI tokens you never spend intopositions that actually run. One loop that puts an idle monthly budget to work on

AgentsLighter CoreLighter Robinhood
Visit

About Lobster Agent Trader

LOBSTER AGENT TRADER

  1. OVERVIEW AND PROBLEM

Developers regularly receive monthly AI token allowances. A significant portion
expires unused, evaporating without rollover or cash-out options. At the same
time, continuous quantitative routines—such as dynamic hedging, rebalancing, and
risk monitoring—are too small to justify manual attention yet too repetitive for
humans to execute reliably. Lobster Agent Trader bridges both gaps. It converts
expiring token balances into disciplined trading loops on Lighter DEX. It is an
open, observable engine offering pause control, audit logs, and rigid deterministic
constraints that language models cannot alter.

  1. TOKEN ALLOWANCE CONTROL

Tradable allowance equals monthly token budget minus reserved work budget plus
carry-over. The reserved balance remains strictly protected against bugs or runaway
loops. Rather than relying on rigid ceilings, the engine projects cycle costs
dynamically using moving averages from recent model calls. Pre-flight checks
evaluate remaining allowances and daily cycle limits prior to invoking the LLM,
ensuring stopped or blocked runs consume zero tokens.

  1. DECISION AND EXECUTION WORKFLOW

Execution proceeds through five distinct stages:

  • Allowance Gate: Evaluates available budget before querying market data APIs.
  • Market Snapshot: Aggregates order books, candles, and native Lighter funding
    rates to prevent misleading cross-exchange carry inversions.
  • Structured Decision: The model generates strict JSON defining symbol, action,
    size, confidence, and invalidation rules. Schema errors undergo two automated
    repair attempts.
  • Execution: Translates dollar amounts into base asset units rounded down to market
    precision, dispatching safe arrays directly to the toolkit.
  • Booking: Logs orders strictly at verified fill prices and quantities, marking
    unfilled orders as blocked to keep records synchronized with exchange balances.
  1. GUARDRAILS AND SECURITY

Risk management enforces strict clamping: oversized requests are capped at preset
limits rather than expanded. Default safeguards include a 250 dollar trade limit,
3x leverage ceiling, 48 daily cycles, 20 dollar daily loss shutdown, and a 0.55
confidence floor.

Transitioning to live trading demands three synchronized safeguards: setting
LIGHTER_ENABLE_LIVE=1, enabling the UI toggle, and manually entering "LIVE". Private
keys reside strictly in file permissions set to 0600, never exposed to databases or
browsers. The service binds exclusively to localhost with constant-time
authentication and cross-site request validation.

  1. PAPER MODE AND RUNTIME

Paper trading operates under cross-margin, fills taker orders instantly, and omits
funding rates. It validates prompt logic and token economics rather than live
profitability. The engine runs on Node 22 with native SQLite and optimizes process
latency through intelligent caching.

Suggestions

Tell the team what to build next. Upvote the ideas you want most - the team uses this to gauge demand.

Posting as preview Ā·

0/1000

No suggestions yet. Be the first to share an idea.

Preview