Skip to content

configuration.md

Latest commit

 

History

History
157 lines (117 loc) · 5.58 KB

configuration.md

File metadata and controls

157 lines (117 loc) · 5.58 KB
Raw

Configuration & Environment

Philosophy

AIMM is designed to run with an LLM token — no LLM-less fallback paths. This eliminates the AI_MARKET_MAKER_USE_LLM toggle. If you have an OPENAI_API_KEY (or compatible endpoint), the system runs at full capability.

Three configuration layers:

.env                        → env vars (secrets, overrides)
config/app.default.json     → app defaults (tuning params, presets)
config/policy.default.json  → trading policy (risk, sizing, rules)

Required Environment Variables

Variable Purpose
OPENAI_API_KEY LLM provider key (OpenAI / compatible)
BINANCE_API_KEY Binance API key for market data
BINANCE_API_SECRET Binance API secret
NEXUS_API_KEY Olaxbt Nexus data API key

No LLM key → the process exits with a clear error. No fallback, no silent degradation. This is intentional — the system is an LLM-native architecture.

Optional Environment Variables

Arbitrator Mode

Variable Values Default
AIMM_ARBITRATOR_MODE weighted_convergence / llm / legacy weighted_convergence
  • weighted_convergence (default): deterministic factor-weighted engine, no LLM cost
  • llm: LLM-based synthesis (uses OPENAI_API_KEY)
  • legacy: original Tier-0 consensus voting

Agent Toggles

Variable Effect
AIMM_ORCHESTRATOR_DISABLE Skip policy orchestrator node
AIMM_TA_TIER0_DISABLE Disable technical TA agent
AIMM_RISK_GUARD_KILL_SWITCH Emergency stop — blocks all trades

Each agent derives its enabled/disabled state from:

  1. Hardcoded default (e.g., Whale 4.1 is disabled by default)
  2. Environment variable override
  3. Orchestrator policy override (when orchestrator is active)

Debug & Observability

Variable Purpose
AIMM_DEBUG_RISK Verbose risk calculation logs
AIMM_FLOW_INCLUDE_FULL_DEBATE Include full debate transcript in output
AIMM_LLM_DESK_DEBATE Enable LLM desk debate (costly)
STRATEGY_INTERVAL_SEC Graph run interval (default: 180)

Execution

Variable Values Default
AI_MARKET_MAKER_ALLOW_LIVE 0 / 1 0
AI_MARKET_MAKER_EXECUTION_ENGINE legacy / oms legacy
MODE paper / live / backtest paper

Opt-In / Opt-Out Architecture

The system is designed as a modular pipeline where each component can be individually enabled or disabled:

policy_orchestrator ── [AIMM_ORCHESTRATOR_DISABLE]
        │
desk_market_scan ── [always on]
        │
├─ monetary_sentinel ── [config weight=0 → skip]
├─ news_narrative_miner ── [config weight=0 → skip]
├─ pattern_recognition_bot ── [config weight=0 → skip]
├─ statistical_alpha_engine ── [config weight=0 → skip]
├─ technical_ta_engine ── [AIMM_TA_TIER0_DISABLE or weight=0]
├─ retail_hype_tracker ── [config weight=0 → skip]
├─ pro_bias_analyst ── [config weight=0 → skip]
├─ whale_behavior_analyst ── [disabled by default in weight config]
├─ liquidity_order_flow ── [config weight=0 → skip]
        │
desk_risk ── [always on]
desk_debate ── [always on, LLM part guarded by AIMM_LLM_DESK_DEBATE]
signal_arbitrator ── [AIMM_ARBITRATOR_MODE selects engine]
portfolio_proposal ── [always on]
desk_risk_guard ── [AIMM_RISK_GUARD_KILL_SWITCH]
portfolio_execute ── [MODE controls real vs paper]
audit ── [always on]

Enabling/disabling agents via weight config

Setting an agent's weight to 0.0 effectively disables it. The remaining enabled weights are re-normalized automatically:

weights = {"2.1": 0.25, "2.3": 0.30, "4.2": 0.15}  # others set to 0
# Normalized internally: 0.25 + 0.30 + 0.15 = 0.70 ≠ 1.0
# Each weight scaled by 1/0.70: 0.357 + 0.429 + 0.214 = 1.0

This is useful for:

  • Minimal mode (Pattern + TA + Liquidity only → 70% of original voting power)
  • Testing (single agent active → verify its factor extraction)
  • Abnormal regimes (disable retail hype during high-vol, etc.)

Layer-Specific Env Design

Layer 0 — Data Sources (none required, but recommended)

BINANCE_API_KEY / SECRET   ← OHLCV data
NEXUS_API_KEY              ← Nexus/Olaxbt data (news, KOL, OI, funding)
TWITTER_BEARER_TOKEN        ← Social sentiment (optional, experimental)

Layer 1 — Agents (optional toggles)

AIMM_TA_TIER0_DISABLE      ← Technical TA

Layer 2 — Arbitrator

AIMM_ARBITRATOR_MODE       ← Engine selection

Layer 3 — Execution

MODE                       ← paper / live / backtest
AI_MARKET_MAKER_ALLOW_LIVE ← double-gate for live

Layer 4 — Safety

AIMM_RISK_GUARD_KILL_SWITCH ← emergency stop