AI Provider Configuration

Overview

NeuraTrade supports multiple AI providers with automatic failover for autonomous trading decisions. The platform uses a provider chain that tries providers in order until one succeeds.

Supported Providers

Anthropic

Claude models (recommended)

OpenAI

GPT-4 and GPT-3.5 models

Zhipu

Chinese AI provider (GLM models)

MiniMax

Alternative Chinese provider

MLX

Local inference (Apple Silicon)

Provider Configuration

Primary Provider

Set your primary AI provider in ~/.neuratrade/config.json:

config.json

{
  "ai": {
    "provider": "anthropic",
    "api_key": "sk-ant-...",
    "base_url": "https://api.anthropic.com/v1"
  }
}

ai.provider

string

default:"zhipu"

Primary AI provider name.Options: anthropic, openai, zhipu, minimax, mlx

ai.api_key

string

required

API key for the primary provider.

Keep your API keys secure. Never commit them to version control.

ai.base_url

string

Custom base URL for the provider API (optional).Defaults:

Anthropic: https://api.anthropic.com/v1
OpenAI: https://api.openai.com/v1
Zhipu: https://open.bigmodel.cn/api/coding/paas/v4
MiniMax: https://api.minimax.chat/v1
MLX: http://localhost:8080/v1

Failover Chain Configuration

Environment Variable

NEURATRADE_AI_PROVIDER_CHAIN

string

default:"zhipu,minimax"

Comma-separated list of fallback providers.The primary provider is automatically added first, followed by the chain.

NEURATRADE_AI_PROVIDER_CHAIN=anthropic,openai,zhipu

Chain Behavior

The failover chain tries providers in order:

Primary Provider (from config.json)
Failover Providers (from NEURATRADE_AI_PROVIDER_CHAIN)

Max Failover Hops

NEURATRADE_AI_FAILOVER_MAX_HOPS

number

default:"1"

Maximum number of failover attempts.

0: No failover (primary only)
1: Try primary + 1 fallback
2: Try primary + 2 fallbacks
-1: Try all providers in chain

NEURATRADE_AI_FAILOVER_MAX_HOPS=2

Provider-Specific Configuration

Anthropic (Claude)

{
  "ai": {
    "provider": "anthropic",
    "api_key": "sk-ant-api03-..."
  }
}

Supported Models:

claude-3-5-sonnet-20241022 (recommended)
claude-3-opus-20240229
claude-3-haiku-20240307

OpenAI

{
  "ai": {
    "provider": "openai",
    "api_key": "sk-proj-..."
  }
}

Supported Models:

gpt-4-turbo-preview
gpt-4-1106-preview
gpt-3.5-turbo-1106

Zhipu (GLM)

{
  "ai": {
    "provider": "zhipu",
    "api_key": "your-zhipu-key"
  }
}

Supported Models:

glm-4
glm-3-turbo

MiniMax

{
  "ai": {
    "provider": "minimax",
    "api_key": "your-minimax-key"
  }
}

MiniMax exposes an Anthropic-compatible API endpoint.

MLX (Local Inference)

{
  "ai": {
    "provider": "mlx",
    "base_url": "http://localhost:8080/v1"
  }
}

Setting up MLX Local Inference

MLX is for local inference on Apple Silicon Macs:

Install MLX: pip install mlx-lm
Download a model: mlx_lm.download --model mistralai/Mistral-7B-v0.1
Start server: mlx_lm.server --model mistralai/Mistral-7B-v0.1 --port 8080
Configure NeuraTrade to use mlx provider

MLX does not require an API key. It runs entirely on your local machine.

Advanced Configuration

Request Timeout

NEURATRADE_AI_HTTP_TIMEOUT_SECONDS

number

default:"300"

HTTP timeout for AI provider requests in seconds.

NEURATRADE_AI_HTTP_TIMEOUT_SECONDS=600

Retry Configuration

NEURATRADE_AI_MAX_RETRIES

number

default:"5"

Maximum number of retries for failed requests.

NEURATRADE_AI_MAX_RETRIES=3

Model Override

Override models for specific providers:

# Override Anthropic model
NEURATRADE_AI_PROVIDER_ANTHROPIC_MODEL=claude-3-opus-20240229

# Override OpenAI model
NEURATRADE_AI_PROVIDER_OPENAI_MODEL=gpt-4-1106-preview

Failover Example

Configuration

.env

# Primary provider
ANTHROPIC_API_KEY=sk-ant-...

# Failover chain
NEURATRADE_AI_PROVIDER_CHAIN=openai,zhipu,minimax

# Fallback API keys
OPENAI_API_KEY=sk-proj-...
ZHIPU_API_KEY=your-zhipu-key
MINIMAX_API_KEY=your-minimax-key

# Max 2 failover attempts
NEURATRADE_AI_FAILOVER_MAX_HOPS=2

Execution Flow

Request Flow:
1. Try Anthropic (primary) → Rate limited (429)
2. Try OpenAI (failover #1) → Success ✓

Failover Stats:
{
  "total_requests": 1,
  "failover_attempts": 1,
  "failover_successes": 1,
  "last_attempt": {
    "primary_provider": "anthropic",
    "attempted_providers": ["anthropic", "openai"],
    "failed_providers": ["anthropic"],
    "success_provider": "openai",
    "failover_attempted": true,
    "failover_succeeded": true
  }
}

Cost Tracking

NeuraTrade tracks AI costs per request:

{
  "usage": {
    "input_tokens": 1250,
    "output_tokens": 430,
    "total_tokens": 1680
  },
  "cost": {
    "input_cost": "0.015",
    "output_cost": "0.0645",
    "total_cost": "0.0795"
  }
}

Budget Limits

AI_DAILY_BUDGET

decimal

default:"10.00"

Maximum daily AI spending in USD.

AI_DAILY_BUDGET=50.00

AI_MONTHLY_BUDGET

decimal

default:"200.00"

Maximum monthly AI spending in USD.

AI_MONTHLY_BUDGET=1000.00

Error Handling

Retryable Errors

Rate limiting (HTTP 429)
Timeout errors
Server errors (HTTP 5xx)
Network connection errors

Non-Retryable Errors

Invalid API key (HTTP 401)
Context length exceeded
Content filtered
Invalid request format

Monitoring

Check AI Status

curl http://localhost:58080/api/v1/ai/status/:userId

Response

{
  "provider": "anthropic",
  "model": "claude-3-5-sonnet-20241022",
  "available": true,
  "failover_enabled": true,
  "failover_chain": ["openai", "zhipu"],
  "last_request": "2026-03-03T10:30:00Z",
  "total_requests": 145,
  "failed_requests": 3,
  "failover_successes": 2
}

Budget Status

curl http://localhost:58080/api/v1/budget/status \
  -H "Authorization: Bearer YOUR_JWT_TOKEN"

Response

{
  "daily": {
    "spent": "7.45",
    "limit": "10.00",
    "remaining": "2.55",
    "percentage": 74.5
  },
  "monthly": {
    "spent": "156.30",
    "limit": "200.00",
    "remaining": "43.70",
    "percentage": 78.15
  }
}

Security Best Practices

Rotate Keys: Change API keys quarterly
Limit Budgets: Set conservative daily/monthly limits
Monitor Usage: Track spending on provider dashboards
Separate Keys: Use different keys for dev/staging/prod
Environment Isolation: Never use production keys in development

Troubleshooting

Provider Not Responding

# Check provider configuration
echo $ANTHROPIC_API_KEY

# Test API key directly
curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model":"claude-3-5-sonnet-20241022","max_tokens":10,"messages":[{"role":"user","content":"test"}]}'

# Check failover stats
curl http://localhost:58080/api/v1/ai/status/your-user-id

Budget Exceeded

# Check current budget
curl http://localhost:58080/api/v1/budget/status

# Increase limits (environment variables)
AI_DAILY_BUDGET=20.00
AI_MONTHLY_BUDGET=400.00

# Or wait for daily reset (midnight UTC)

Failover Not Working

# Verify failover chain
echo $NEURATRADE_AI_PROVIDER_CHAIN

# Check max hops
echo $NEURATRADE_AI_FAILOVER_MAX_HOPS

# Verify fallback API keys are set
env | grep API_KEY

Get Started

Core Features

Configuration

Guides

​Overview

​Supported Providers

Anthropic

OpenAI

Zhipu

MiniMax

MLX

​Provider Configuration

​Primary Provider

​Failover Chain Configuration

​Environment Variable

​Chain Behavior

​Max Failover Hops

​Provider-Specific Configuration

​Anthropic (Claude)

​OpenAI

​Zhipu (GLM)

​MiniMax

​MLX (Local Inference)

​Advanced Configuration

​Request Timeout

​Retry Configuration

​Model Override

​Failover Example

​Configuration

​Execution Flow

​Cost Tracking

​Budget Limits

​Error Handling

​Retryable Errors

​Non-Retryable Errors

​Monitoring

​Check AI Status

​Budget Status

​Security Best Practices

​Troubleshooting

Build docs developers (and LLMs) love

Overview

Supported Providers

Provider Configuration

Primary Provider

Failover Chain Configuration

Environment Variable

Chain Behavior

Max Failover Hops

Provider-Specific Configuration

Anthropic (Claude)

OpenAI

Zhipu (GLM)

MiniMax

MLX (Local Inference)

Advanced Configuration

Request Timeout

Retry Configuration

Model Override

Failover Example

Configuration

Execution Flow

Cost Tracking

Budget Limits

Error Handling

Retryable Errors

Non-Retryable Errors

Monitoring

Check AI Status

Budget Status

Security Best Practices

Troubleshooting