Free and open source. The AI reads your Form 16 and AIS and interviews you for missed deductions; a tested Python engine does all the math. No CA, no ₹3,000 fee, and only you ever click Pay, Submit, or e-Verify.
Still filing for FY 2025-26? That works. Planning FY 2026-27 advance tax? Start with your expected annual income and TDS. It confirms your year and task first.
/plugin marketplace add karanb192/itr-wala
codex plugin marketplace add https://github.com/karanb192/itr-wala
[NEW REGIME] Total income 25,06,700 NET PAYABLE (-ve=refund) 720 [OLD REGIME] Total income 18,74,300 NET PAYABLE (-ve=refund) 43,530 RECOMMENDED: NEW regime (saves Rs. 42,811)
python3 skills/itr-wala/scripts/tax_engine.py skills/itr-wala/assets/example-income.jsonEvery AI tax demo this season had the same silent flaw: the model was doing the arithmetic. LLMs are magnificent at reading a Form 16 and terrible at applying s.87A marginal relief. itr-wala splits the work the way it should be split.
Used itr-wala before 1.3.0? Plugins 1.2.0 and 1.2.1 and older clone/skills.sh installs are affected. For FY 2025-26 new-regime computations, recheck special-rate gains or winnings when total income exceeded ₹12 lakh and the old output granted rebate or marginal relief, including slab income above ₹12 lakh. Old computation JSON identifies the engine as 1.2.0. Check the affected examples and next steps.
Version 1.3.1 corrects more edge cases. Both supported years receive fixes for home-loan caps, late fees, business elections, capital-gain loss allocation, NPS, medical claims and slab rounding. Check whether an older result needs another look.
These layers catch real bugs. The fuzzer found a one-in-350,000 rounding edge where ₹52,880 more salary computed ₹10 less tax. Fixed, pinned as a regression test. That find-fix-pin loop is the thing a prompt-only tax tool cannot run. If your CA can show you their test suite, hire them.
Self-test first, so the engine proves its math before touching your return. Then documents, extraction with validation, a deduction hunt, both regimes computed, and a filing pack that maps every portal field to its value.
Historical FY 2025-26 recording from July 2026. It shows the earlier 47-test build; the current workflow confirms your year and task first.
Out of scope, and it says so instead of guessing: non-residents, F&O and intraday, audit cases, foreign tax credit, ESOP deferral. Built by someone who files his own taxes with it and does security for a living.
No, and that is the point. You are trusting a tested Python engine with the math and an LLM with reading PDFs and explaining things, which are the two things each is actually good at. The test suite runs in front of you before your return is touched, and you can run it yourself.
Yes. You prepare your own return and file it yourself on the government portal, same as using the portal's own forms, just with better preparation. The tool never submits anything on your behalf.
Nothing. MIT-licensed, free forever. The only cost is your existing Claude Code (or Codex/Gemini CLI) subscription doing the document reading. The tax math is plain Python and costs nothing no matter how many times you recheck.
Use whatever you trust. This is for people who would rather review every number themselves than pay ₹3,000+ to hope someone else did. The filing pack it generates is also a great ₹0 first draft to hand a CA for a cheap review, it is built to be handed over.
Resident individuals for FY 2025-26 and FY 2026-27: supported salary, house property, capital gains, crypto, winnings, interest, family pension and presumptive income, with both regimes and advance-tax planning. FY 2025-26 keeps ITR-1 through ITR-4 and belated filing guidance. FY 2026-27 uses Tax Year 2026-27 under the 2025 Act; notified forms and utility must be verified before final filing guidance. Supported house-property loss set-off is included. Loss carry-forwards, buyback capital-loss entries, FY 2026-27 buybacks and SGB exemption classification need separate review. Full scope.