Open Source

Personal finance engine for AI agents.

Verified federal + state data with deterministic tax, retirement, and estate calculations — local by default, sub-ms, JSON-in/JSON-out.

curl -fsSL https://get.entropyfa.com | sh
Read the Docs View on GitHub

See it in action

entropyfa 
# What data is available?
$ entropyfa data coverage
{"data":{"count":17,"entries":[{"category":"tax","key":"brackets",...},...]},"ok":true}

# Look up 2026 tax brackets
$ entropyfa data lookup --category tax --key brackets --filing-status single
{"data":[{"rate":0.10,"min":0,"max":12400},{"rate":0.12,...},...],ok":true}

# Optionally POST the same envelope to a webhook
$ entropyfa --result-hook-url https://example.com/hook data lookup --category tax --key brackets --filing-status single

# Compute federal tax
$ entropyfa compute federal-tax --json '{"filing_status":"single","income":{"wages":150000}}'
{"data":{"total_tax":36209,"effective_rate":0.1649,"marginal_ordinary_rate":0.24,...},"ok":true}

# What does this command need?
$ entropyfa compute roth-conversion --schema
{"data":{"command":"roth-conversion","gather_from_user":{"required":[...]},...},"ok":true}

Why We Built This

The world is moving to agents. Personal finance will too. But financial planning agents need two things they can't do well on their own:

Verified data — rates, limits, rules, and tables that change annually. Must be IRS-sourced, not hallucinated.

Deterministic math — tax bracket stacking, actuarial calculations, Monte Carlo simulations, portfolio management. Same input, same output, every time.

entropyfa bundles both into a single binary with zero configuration.

Verified IRS Data

Every rate, bracket, table, and threshold sourced directly from IRS publications. No hallucination, no stale training data.

Sub-ms Deterministic

Pure computation with zero I/O. Same input always produces the same output. Agents can reason about results with confidence.

Zero Config / Single Binary

One curl command to install. No API keys, no config files, and no outbound calls unless you explicitly use upgrade or result hooks.

Agent-Native

--schema on every command tells agents what to gather. JSON-in/JSON-out. Works with Claude, GPT, LangChain, or plain shell.

What's Inside

Data

  • • Tax brackets (by filing status)
  • • Standard deductions
  • • Capital gains brackets
  • • Estate tax exemption & brackets
  • • NIIT thresholds
  • • Payroll rates (SS, Medicare)
  • • QBI deduction parameters
  • • RMD tables (Uniform, Single, Joint Life)
  • • IRMAA brackets
  • • SS taxation thresholds
  • • 417(e) mortality tables

Compute

  • federal-tax — Federal income + payroll taxes
  • estate-tax — Federal estate tax (Form 706)
  • rmd — Required minimum distribution
  • rmd-schedule — Multi-year RMD projection
  • roth-conversion — Conversion tax impact
  • roth-conversion-strategy — Multi-year strategy
  • pension-comparison — Lump sum vs annuity
  • projection — Monte Carlo / linear projection
  • goal-solver — Binary search goal solver

Designed for Agents

entropyfa is built as a tool for AI agents doing financial planning.

›

--schema on every command — agents read the schema to know what inputs to gather

›

data coverage — agents discover available reference data without hardcoding

›

JSON-in/JSON-out — structured I/O that agents parse natively

›

Deterministic — same input always produces the same output

›

No configuration — install and go, no API keys, no config files

Works with any agent framework — Claude tool use, OpenAI function calling, LangChain, or plain shell exec.