LLM Tools¶

LLM-powered data transformation tools. Takes input data and a prompt, uses an LLM to process/transform it.

Highlights¶

Single function for any data transformation
Configurable model via OpenRouter or OpenAI-compatible API
Chain with other tools for structured output extraction

Functions¶

Function	Description
`llm.transform(data, prompt, ...)`	Transform data using LLM instructions
`llm.transform_file(prompt, in_file, out_file, ...)`	Transform file content and write to output

Key Parameters¶

Parameter	Type	Description
`data`	any	Data to transform (converted to string)
`prompt`	str	Instructions for transformation
`in_file`	str	Input file path (for transform_file)
`out_file`	str	Output file path (for transform_file)
`model`	str	AI model to use (uses `transform.model` from config)
`json_mode`	bool	If True, request JSON output format (default: False)

Requires¶

Configuration (tool not available until all are set): - OPENAI_API_KEY in secrets.yaml - transform.base_url in onetool.yaml (e.g., https://openrouter.ai/api/v1) - transform.model in onetool.yaml (e.g., openai/gpt-5-mini)

Examples¶

# Extract structured data from search results
llm.transform(
    data=brave.search(query="gold price today"),
    prompt="Extract the current gold price in USD/oz as a single number"
)

# Convert to YAML format
llm.transform(
    data=some_data,
    prompt="Return ONLY valid YAML with fields: name, price, url"
)

# Summarize content
llm.transform(
    data=web.fetch(url="https://example.com/article"),
    prompt="Summarize the main points in 3 bullet points"
)

# Get JSON output with json_mode
llm.transform(
    data=raw_data,
    prompt="Extract name and email fields",
    json_mode=True
)

# Transform a file
llm.transform_file(
    prompt="Convert this markdown to reStructuredText format",
    in_file="README.md",
    out_file="README.rst"
)

Configuration¶

Add to onetool.yaml:

tools:
  transform:
    base_url: https://openrouter.ai/api/v1
    model: openai/gpt-5-mini
    timeout: 30        # API timeout in seconds (default: 30)
    max_tokens: 4096   # Maximum tokens in response (optional)