LLM Node

The LLM node calls AI language models to analyze data, generate insights, and make decisions within your workflow. It supports 9 providers and 30+ models — from fast summarizers to deep reasoning engines.

Configuration

Available Models

Model Selection Guide

Not sure which model to pick? Match the use case to the right category:

Prompt Interpolation

Use double curly braces to inject live data from upstream nodes into your prompts.

Example Prompts

Market Analyst:

System: You are a crypto market analyst. Analyze the provided price data
and technical indicators.

Respond in this exact format:
ACTION: [BUY / SELL / HOLD]
CONFIDENCE: [0-100]%
RATIONALE: [2-3 sentence explanation]

Be conservative — only recommend BUY when multiple indicators align.

User: Analyze BTC/USD right now.

Current price: {{price_data.data.prices[0].current}}
24h change: {{price_data.data.prices[0].changePercent24h}}%
RSI: {{price_data.data.prices[0].indicators.rsi}}

Based on these indicators, what is your recommendation?

Risk Monitor:

System: You are a portfolio risk manager. Review the positions and flag
any concerns.

Respond in valid JSON only:
{"alerts": [{"symbol": "...", "issue": "...", "severity": "low|medium|high"}],
 "overallRisk": "low|medium|high",
 "summary": "..."}

User: Review my current positions:
{{portfolio.positions}}

Cash balance: {{portfolio.cashBalance}}
Net worth: {{portfolio.netWorth}}

Prediction Market Analyst:

System: You are a prediction market analyst. Given the market title,
current probability, and description, assess whether the current
probability represents good value.

Respond in this format:
POSITION: [BUY_YES / BUY_NO / SKIP]
EDGE: [your estimated probability minus market probability]%
REASONING: [1-2 sentences]

User: Evaluate this Polymarket market:
Title: {{polymarket.markets[0].title}}
Probability: {{polymarket.markets[0].markets[0].outcomePrices[0]}}
Description: {{polymarket.markets[0].description}}

Structured Output

Toggle Structured Output to force the model to return a specific JSON schema instead of free-form text. This is useful when you need to feed parsed data directly into Conditional or Function nodes without extra parsing.

Define fields with a name, type (string, number, boolean, array, or object), and whether they're required. Click the type badge to change it. Expand objects and arrays to add nested properties. Try it below — the output preview updates in real time.

{
  "signal": "buy",
  "confidence": 0.85,
  "reasoning": "RSI below 30 indicates oversold"
}

The demo above is pre-populated with a trading signal schema. In your workflow, the LLM will return JSON matching your schema exactly — you can then route it directly into a Conditional node (e.g., check if signal equals "buy" and confidence is greater than 0.7).

Visual Analysis

Connect a Chart Image node to the LLM to enable visual chart analysis. Vision-capable models (GPT-4o, Gemini, Kimi K2.5) can analyze candlestick patterns, support/resistance levels, and trend direction directly from the chart image.

The LLM automatically detects images in the interpolated data — no special configuration needed.

Parsing LLM Output Downstream

The LLM returns a text string by default. To use it in decisions:

• Simple routing: Connect LLM → Conditional. Set Field to llm.output, Operator to "Contains", Value to BUY. The true branch triggers the trade, false branch sends a notification.

• Structured parsing: Enable Structured Output on the LLM node itself, or connect LLM → Function node that parses the text into JSON → Conditional on the parsed fields.

• Multi-model consensus: Run the same data through multiple LLM nodes in parallel, then merge results in a Function node to vote on the final action.

Pricing & Credits

When you use NickAI Credits (the default), each LLM call costs credits based on what the provider charges us. The credit cost depends on:

• Which model you choose — larger, more capable models cost more per token than lightweight ones.
• How many tokens the request uses — both the input (your prompts + interpolated data) and the output (the model's response) count toward the total.

Longer prompts, bigger context windows, and higher maxTokens values all increase the cost. If you want to minimize credit usage, choose a smaller model (e.g., Claude Haiku 4.5, GPT-4o Mini) and keep your prompts concise.

Output

Next Steps

• Conditional Node — Route decisions based on LLM output.
• Function Node — Parse or transform LLM responses with custom code.
• Chart Image Node — Generate charts for visual LLM analysis.
• Credentials — Set up your own API keys for each provider.