Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.tracia.io/llms.txt

Use this file to discover all available pages before exploring further.

Use variables to inject dynamic content into your messages with {{placeholder}} syntax.

​
Basic Usage

result = client.run_local(
    model="gpt-4o",
    messages=[
        {"role": "system", "content": "You are a {{role}}."},
        {"role": "user", "content": "Help me with {{task}}."},
    ],
    variables={
        "role": "helpful coding assistant",
        "task": "writing a REST API",
    },
)
The messages sent to the LLM will be:
  • System: “You are a helpful coding assistant.”
  • User: “Help me with writing a REST API.”

​
Variable Syntax

Variables use double curly braces: {{variableName}}
  • Variable names must contain only alphanumeric characters and underscores
  • Variables are case-sensitive
  • Undefined variables are left as-is (not replaced)
result = client.run_local(
    model="gpt-4o",
    messages=[
        {"role": "user", "content": "Hello {{name}}, how is {{city}}?"},
    ],
    variables={
        "name": "Alice",
        # "city" is not provided
    },
)
# Result: "Hello Alice, how is {{city}}?"

​
Multi-Message Interpolation

Variables are interpolated across all messages: 
result = client.run_local(
    model="gpt-4o",
    messages=[
        {"role": "system", "content": "You are an expert in {{language}}."},
        {"role": "user", "content": "How do I create a class in {{language}}?"},
        {"role": "assistant", "content": "In {{language}}, you create a class using..."},
        {"role": "user", "content": "Can you show me an example in {{language}}?"},
    ],
    variables={"language": "Python"},
)

​
Common Patterns

​
Personalized Responses

result = client.run_local(
    model="gpt-4o",
    messages=[
        {"role": "system", "content": "You are helping {{user_name}} with their {{project_type}} project."},
        {"role": "user", "content": "{{user_question}}"},
    ],
    variables={
        "user_name": "Alice",
        "project_type": "e-commerce",
        "user_question": "How should I structure the database?",
    },
)

​
Dynamic Prompts from Configuration

prompt_config = {
    "system_prompt": "You are a {{tone}} assistant specializing in {{domain}}.",
    "user_template": "Please {{action}} the following: {{content}}",
}

result = client.run_local(
    model="gpt-4o",
    messages=[
        {"role": "system", "content": prompt_config["system_prompt"]},
        {"role": "user", "content": prompt_config["user_template"]},
    ],
    variables={
        "tone": "friendly",
        "domain": "web development",
        "action": "review",
        "content": user_input,
    },
)

​
Locale-Based Content

translations = {
    "en": {"greeting": "Hello", "farewell": "Goodbye"},
    "es": {"greeting": "Hola", "farewell": "Adiós"},
    "fr": {"greeting": "Bonjour", "farewell": "Au revoir"},
}

locale = "es"
result = client.run_local(
    model="gpt-4o",
    messages=[
        {"role": "system", "content": 'Respond using "{{greeting}}" and "{{farewell}}" appropriately.'},
        {"role": "user", "content": "Greet the user and say bye."},
    ],
    variables=translations[locale],
)

​
Variables vs f-strings

You might wonder why use Tracia variables instead of Python f-strings. The key difference is traceability. 
result = client.run_local(
    model="gpt-4o",
    messages=[
        {"role": "user", "content": "Summarize {{topic}} for {{audience}}."},
    ],
    variables={
        "topic": "machine learning",
        "audience": "beginners",
    },
)
# Span stores: interpolated messages + original variables
# Variables are preserved for filtering and analysis

​
With f-strings

topic = "machine learning"
audience = "beginners"

result = client.run_local(
    model="gpt-4o",
    messages=[
        {"role": "user", "content": f"Summarize {topic} for {audience}."},
    ],
)
# Span stores: only the final string
# No way to filter by variable values
Using Tracia variables allows you to:
  • Filter spans by variable values in the dashboard
  • Analyze prompt performance across different inputs
  • Group spans by the variables used

​
Escaping Curly Braces

If you need literal curly braces in your content, they won’t be interpolated if they don’t match the {{word}} pattern: 
result = client.run_local(
    model="gpt-4o",
    messages=[
        {"role": "user", "content": "In Python, dicts use {key: value} syntax. My name is {{name}}."},
    ],
    variables={"name": "Alice"},
)
# Result: "In Python, dicts use {key: value} syntax. My name is Alice."
Single braces and braces with spaces inside are preserved as-is.