Wolfram Alpha (LLM API)

Delegate precise, formalizable computations and factual lookups to Wolfram|Alpha via its LLM API (HTTP) to get verified results and reduce arithmetic/modeling errors. Use for numeric calculations, unit conversions, symbolic algebra/calculus, equation solving, statistics/probability, date/time math, finance/math with currency, and structured data queries (e.g., population, chemistry/physics constants, astronomy facts). Use when you want short, LLM-friendly text output (optionally within a character budget), need to resolve ambiguous interpretations via assumptions, or want location/units/language-specific results.

Wolfram|Alpha (LLM API) skill

Use the bundled wrapper script to call Wolfram|Alpha's LLM API and return concise, model-ingestible results.

Preconditions

• Environment variable WOLFRAM_APP_ID must be set (your Wolfram|Alpha AppID). If it is not set, ask your human to set it (do not guess or hardcode keys).

Quick start

Run:

# default: cache ON (7d), auth via bearer header (keeps AppID out of URL)
python3 skills/wolfram-alpha-llm/scripts/wa_llm.py \
  --input "solve x^2 + 3x + 2 = 0"

What to send as --input

• Prefer short English keyword-style queries when possible.
• If the user asked in another language, translate to English for the API call, then answer in the user’s original language.
• When you need an exact computation, be explicit (e.g., integrate sin(x)^2 from 0 to pi).

Core parameters (use these most)

• --input (required): the query.
• --maxchars (optional, default 2500): cap response length.
• --units (optional): set units system, if needed for conversions/physics (metric is often a good default when unspecified).
• --assumption (optional, repeatable): disambiguate when WA returns irrelevant interpretation or offers assumptions.

High-value optional parameters (use when relevant)

• Localization / context:
  • --countrycode, --languagecode
  • --timezone
  • One of: --ip | --latlong | --location (pick exactly one)
• Finance:
  • --currency (e.g., USD, EUR)
• Performance / robustness:
  • --scantimeout, --parsetimeout, --formattimeout, --totaltimeout

Output handling guidance

• Treat output as computed evidence: quote the key result, then add minimal interpretation.
• If the result is too long/noisy, rerun with a smaller --maxchars.
  • Heuristic: for simple conversions / arithmetic / single-value answers, try --maxchars 800.
  • Keep default --maxchars 2500 for most multi-line or explanation-heavy results.
• If the interpretation is wrong:
  1. retry with --assumption ... (use WA-provided suggestions when available),
  2. only then rephrase/simplify --input.

Wrapper script

• Script: skills/wolfram-alpha-llm/scripts/wa_llm.py
• Auth:
  • default --auth bearer: sends Authorization: Bearer <AppID> header (keeps AppID out of the URL)
  • --auth query: sends appid as URL parameter
• Cache:
  • default --cache on with --cache-ttl 604800 (7d)
  • stores best-effort results in: ~/.cache/openclaw-wolfram-alpha/
• Returns:
  • stdout: API text body
  • stderr: errors, HTTP status context

For parameter details and error behaviors, see:

• skills/wolfram-alpha-llm/references/llm-api.md
• skills/wolfram-alpha-llm/references/full-api-params.md