Skip to main content

​
AI Agents API

AI agents are autonomous entities that execute trading decisions using LLM-powered reasoning. NeuraTrade supports three primary agent roles:

​
Agent Roles

​
Analyst Agent

Analyzes market conditions and generates trading signals. Capabilities:
  • Technical indicator analysis
  • Market regime detection
  • Correlation analysis
  • Sentiment aggregation
  • Forecast generation
Configuration: 
{
  role: "analyst",
  provider: "zhipu",
  model: "glm-4-flash",
  context_window: 128000,
  skills: ["technical_analysis", "sentiment_analysis"]
}

​
Trader Agent

Executes trades based on signals and market conditions. Capabilities:
  • Order execution planning
  • Position sizing
  • Entry/exit timing
  • Slippage management
  • Multi-exchange routing
Configuration: 
{
  role: "trader",
  provider: "minimax",
  model: "abab6.5s-chat",
  max_position_size: "500.00",
  max_concurrent_positions: 3,
  leverage: 5
}

​
Risk Manager Agent

Monitors portfolio safety and enforces risk limits. Capabilities:
  • Real-time drawdown monitoring
  • Position size throttling
  • Daily/monthly loss tracking
  • Emergency liquidation
  • Safety gate enforcement
Configuration: 
{
  role: "risk_manager",
  max_daily_loss: "2.00%",
  max_drawdown: "5.00%",
  position_throttle: "100%",
  consecutive_loss_halt: 3
}

​
Get AI Status

curl -X GET "https://api.neuratrade.ai/api/v1/telegram/internal/ai/status/123456789" \
  -H "X-Admin-API-Key: YOUR_ADMIN_KEY"

{
  "chat_id": "123456789",
  "enabled": true,
  "provider": "zhipu",
  "model": "glm-4-flash",
  "failover_chain": ["zhipu", "minimax"],
  "agents": [
    {
      "role": "analyst",
      "status": "active",
      "last_action": "2026-03-03T10:28:45Z"
    },
    {
      "role": "trader",
      "status": "active",
      "last_action": "2026-03-03T10:29:12Z"
    },
    {
      "role": "risk_manager",
      "status": "active",
      "last_action": "2026-03-03T10:30:00Z"
    }
  ],
  "usage": {
    "requests_today": 142,
    "tokens_today": 45230,
    "cost_today": "0.23"
  }
}
Retrieves AI agent status and usage metrics for a chat.

​
Path Parameters

​
chatId
string
required
Telegram chat ID

​
Response Fields

​
enabled
boolean
Whether AI agents are enabled for this chat
​
provider
string
Primary LLM provider: zhipu, minimax, anthropic, mlx
​
model
string
Currently active model name
​
failover_chain
array
Ordered list of provider fallbacks
​
agents
array
Active agent instances
​
usage
object
Usage metrics

​
Agent Configuration

Agents are configured via environment variables and operator settings:

​
Environment Variables

# Primary AI Provider
NEURATRADE_AI_PROVIDER="zhipu"
ZHIPU_API_KEY="your_api_key_here"

# Failover Chain
NEURATRADE_AI_PROVIDER_CHAIN="zhipu,minimax"
MINIMAX_API_KEY="your_minimax_key"

# Timeouts and Retries
NEURATRADE_AI_HTTP_TIMEOUT_SECONDS=300
NEURATRADE_AI_MAX_RETRIES=5
NEURATRADE_AI_FAILOVER_MAX_HOPS=1

# Model Overrides
NEURATRADE_AI_PROVIDER_ZHIPU_MODEL="glm-4-flash"
NEURATRADE_AI_PROVIDER_MINIMAX_MODEL="abab6.5s-chat"

​
Supported Providers

ProviderModelsContextSpeed
Zhipuglm-4-flash, glm-4-plus128KFast
MiniMaxabab6.5s-chat, abab6.5g-chat245KMedium
Anthropicclaude-3.5-sonnet200KFast
MLXLocal modelsVariesFast

​
Agent Execution Flow

Agents operate within quest execution cycles. Each quest invokes the appropriate agent based on task type.

​
Agent Coordination

Agents coordinate through the integrated quest handler system:

​
Scalping Flow

  1. Market Scanner Quest (every 1-5 min)
    • Analyst agent evaluates opportunities
    • Returns top-ranked symbols
  2. Scalping Execution Quest (on signal)
    • Trader agent plans entry
    • Risk manager validates position size
    • Trade executed if approved
  3. Stop-Loss Monitor (continuous)
    • Risk manager tracks open positions
    • Auto-liquidates if stop-loss hit

​
Arbitrage Flow

  1. Funding Rate Scanner (every 5 min)
    • Analyst detects funding rate anomalies
  2. Arbitrage Calculator
    • Trader computes optimal position sizes
    • Risk manager validates capital allocation
  3. Execution
    • Simultaneous spot + futures orders
    • Position tracked until funding settlement
All agent actions are subject to portfolio safety gates. If drawdown exceeds limits or daily loss cap is hit, agents will halt trading until safety conditions are restored.

​
Agent State Persistence

Agent decisions and outcomes are persisted in:
  • Trade Memory (ai_trade_memory table)
    • Entry/exit reasoning
    • Outcome labels (win/loss)
    • Pattern recognition for learning
  • Lifecycle Store (trading_lifecycle_* tables)
    • Order placement history
    • Position tracking
    • Realized PnL
  • Quest Checkpoints
    • In-progress state
    • Retry attempts
    • Error recovery data

Build docs developers (and LLMs) love

Get started for freeTalk to us