v0.2.0MIT · open source

Deterministic layers for Financial Planning

traj is a free, open-source financial simulation tool built for AI agents. Ship as a skill and CLI — local-first, no accounts, no dashboards, no subscriptions.

Install the CLI directly
Before
  1. You

    When can I stop working?

  2. Agent

    Probably around 2040, if markets cooperate.

  3. You

    Try again.

  4. Agent

    Closer to 2046. Hard to say.

  5. You

    How did you come up with that? Summarise it for me.

  6. Agent

    Rule of thumb: 7% a year on your investments, and you draw 4% a year. That lands near 2045.

  7. You

    Are you sure?

  8. Agent

    No. I have none of your numbers, and my data stops in 2024.

Made up, then admitted when asked. No source, no date, nothing to check.

After
  1. You

    When can I stop working?

  2. Agent

    Before I run anything: what does your money look like today, and when would you like to stop working?

  3. You

    Here is the budget sheet I keep.

    budget-2026.csv
  4. Agent

    This covers your spending, but not what you have saved. What is your balance today?

  5. You

    I have 1.2M saved today and I put away 25,000 a month.

  6. Agent

    The agent loads the traj skill and runs the CLI:

    traj fi --scenario base --as-of 2026-03-31 --horizon 2066-03-31 --json
    {  "asOf": "2026-03-31",  "tax": "not_applied",  "method": "survival",  "fi": { "reachable": true, "fiDate": "2041-08-01" }}

    Example output, abridged.

  7. Agent

    You can stop working in August 2041.

    traj fi · as of 31 Mar 2026

The figure came from the CLI, with the date it was computed. Nothing invented.

How it works

Agents own the presentation. traj owns the logic.

You talk to your agent. It uses skills to run traj under the hood, then shows you the result however you like.

  1. 01You ask your agent about your money.You + your agent
  2. 02The agent loads a traj skill.Your agent
  3. 03The skill runs the traj CLI under the hood.The skill + the CLI
  4. 04traj returns exact figures. The agent presents them.traj → your agent
Presentation layerOwned by your agent

Your agent decides how to explain it: the words, the charts, the language.

  • AnswerWhen can I stop working?August 2041.
  • Interactive chart2041
  • ActionsOn trackGoogle Tasks

SkillsSkills teach the agent which commands to run and how to read the answer.

Logic layerOwned by traj

traj does all the arithmetic. Same input, same answer, no model in the loop.

  • Projections
  • Tax
  • FI date
  • Ledger

Features

Made for agents, built to be checked

  • 01DeterministicSame input, same answer. The CLI never calls a model, so every result is repeatable.
  • 02Agent-operatedEvery command describes its own input and output with traj schema, so an agent never guesses a flag.
  • 03Local-firstRuns on your machine. No accounts, no dashboards, no subscriptions.
  • 04Open sourceFree and MIT licensed. Read the code that computes your future.
  • 05Tax on requestTax is applied only when you ask for it, and every figure says whether it includes tax.
  • 06Shows its workingEvery projection, comparison and FI result states its as-of date, engine version and scenario in its JSON.