india-tax-guru 0.2.1

Open-source India income-tax planning, salary-structuring, and ITR-filing-support toolkit

india-tax-guru

Open-source India income-tax planning, salary-structuring, and ITR-filing support toolkit. Library + CLI, usable standalone or as an agent skill (Claude Code SKILL.md, or any AGENTS.md-compatible agent).

npx skills add anivar/india-tax-guru

Not a filing agent for the government e-filing portal, and not a substitute for a CA. Rules are versioned per assessment year and non-obvious edge cases are documented in the module that handles them.

Verify before you file. Tax law is intricate and changes every Budget. Treat the output as a well-tested second opinion, not as authority — check it against the ITD utility or a CA before filing, and please open an issue if a figure disagrees.

Features

• Old vs new regime comparison — full computation under both regimes, always both, never just the one you assumed. Ties go to the new regime, which is the statutory default and needs no Form 10-IEA opt-out.
• Age-aware slabs — the old regime's basic exemption rises to ₹3,00,000 at 60 and ₹5,00,000 at 80, and super-seniors lose the 5% bracket entirely.
• Salary & HRA — period-wise HRA exemption that handles a mid-year rent or city change correctly, plus s.16 standard deduction and professional tax.
• Capital gains — s.111A/112A equity rates, the ₹1,25,000 s.112A exemption, s.50AA (specified mutual funds deemed short-term at slab rates), s.70/74 set-off ordering, and the resident basic-exemption set-off against gains.
• Foreign-listed equity — a US-listed RSU is not a s.112A asset. No STT, not a recognised Indian exchange, so no ₹1,25,000 exemption, no concessional rate, and a 24-month long-term threshold rather than 12. Held as its own asset class precisely because picking equity_listed for it would silently grant all three.
• Presumptive taxation — s.44AD and s.44ADA, including the split rate (6% applies only to the digitally-received slice of turnover, 8% to the rest — it is not an alternative rate on the whole), the ₹3 crore / ₹75 lakh enhanced caps and their 5% cash test, the s.44AD(4) lock-in, and the single 15 March advance-tax instalment.
• GST-turnover reconciliation — for a GST-registered presumptive filer, the ITR turnover is reconciled against the GSTR taxable value the way the AIS cross-match does it, and the two silent failure modes are named: a GST-inclusive turnover (which overpays tax while producing a clean-looking AIS match) and an under-reported one (which files clean and surfaces as an e-verification notice). Exempt, non-GST and pre-registration receipts are declared, not guessed.
• House property — s.24(b) and s.71(3A) caps applied in aggregate across properties, let-out NAV/30%/interest, carry-forward reported not dropped.
• Surcharge — thresholds, the new regime's 25% ceiling, the 15% cap on surcharge attributable to capital gains and dividends, and marginal relief computed against tax recomputed at the threshold.
• Deductions — 80C/80CCD(1B)/80CCD(2)/80D/80TTA vs 80TTB/80DDB/80E/80G, capped, regime-gated, and bounded by gross total income.
• Settlement — TDS, advance tax and self-assessment tax netted off to a refund or balance-payable figure, with s.234B/234C interest.
• CTC / salary restructuring optimizer — searches Basic/HRA/employer-NPS splits within realistic employer policy bounds, across both regimes, to maximize take-home pay.
• Salary-slip analysis — classify and reconcile monthly payslips against Form 16, flagging unclassified line items instead of guessing, and feed them straight into the engine.
• Salary-structure advisory — what to change and what each change is worth, with every figure measured by re-running the engine on the counterfactual rather than estimated from a marginal rate.
• Regime-choice compliance — Form 10-IEA only where business income makes it necessary, and the s.139(1) deadline that decides whether the old regime is available at all.
• HUF assessees — the individual machinery minus the reliefs an HUF cannot claim: no s.87A rebate, no age-based exemption, no salary heads, no 80CCD(1B) or 80E, and no s.44ADA (individuals and firms only since Finance Act 2021). Illegal inputs — an HUF with salary income, a senior age band, or an NPS deduction — are rejected at construction rather than silently taxed.

Not implemented — and why

• Anyone but an individual or HUF. AOP/BOI, firm, LLP and company are each taxed under different rules — a firm pays a flat 30% with no regime choice; a company sits outside s.115BAC entirely, with its own rates and minimum alternate tax. Constructing a profile with any of these raises UnsupportedAssesseeError rather than quietly handing back an individual's tax, because that wrong answer would look completely ordinary.
• Actual (non-presumptive) business income, and the tax-audit machinery.
• Foreign income and foreign assets: Schedule FA disclosure, DTAA relief and foreign tax credit, and the vesting-stage perquisite taxation of RSUs/ESOPs. (The sale gain of foreign-listed stock IS modelled — see foreign_equity above; it is everything around it that is not.)
• Pre-construction home-loan interest amortisation.
• Capital-loss carry-forward across years — this is a single-year computation.
• s.234C's carve-out for gains arising after an instalment due date, which needs transaction-level dates.
• Direct e-filing or portal automation.

Document importers: the agent is the importer

There is no code that parses the prefill JSON, AIS JSON or Form 16 Part B PDF, and that is a decision rather than a gap. ITD publishes no schema for any of them. Everything that could be learned about their key names came from a single real file each — n=1 — and the one Form 16 sample already contained two different Part B renderings with shifted sub-item letters and a self-inconsistent 80D figure. A parser built on that would work on one payroll vendor and fail silently on the rest, and silent failure in a tax tool is the failure mode this project exists to avoid.

What works instead is to let an agent do the extraction and this engine do the arithmetic. Reading a semi-structured document whose exact shape it has never seen is what a model is good at; arithmetic nobody can unit-test is what it is bad at. So the agent fills in profile.json and never states a tax figure of its own. docs/importing.md carries the per-document guidance and the reconciliation checks to run before trusting any extracted value.

Two things are worth building and are the natural next step, because they rest on published or self-describing formats rather than guesswork:

• The ITR export schemas are real JSON Schema documents published by ITD, with additionalProperties: false set at the root. Generate types from them and validate in CI; do not hand-write the structs.
• Form 26AS in its text export is caret-delimited and carries its own column headers per part, so it can be parsed as a self-describing stream rather than against a fixed layout.

Every limitation above is also called out at the point in the code where it would otherwise silently produce a wrong number — grep for not modelled for the full list with context.

Why per-year rule modules

Tax law changes every Budget. Rather than one code path with if year >= 2024 branches accumulating forever, each assessment year is a small, self-contained module under src/india_tax_guru/rules/ with the source Finance Act cited in a comment. Adding a new year means writing a new file, not editing old ones — so last year's numbers can never regress.

Install

As an agent skill (Claude Code or any AGENTS.md-compatible agent) — the main way to use this:

npx skills add anivar/india-tax-guru

As a CLI, installed once and on your PATH:

uv tool install "india-tax-guru[cli] @ git+https://github.com/anivar/india-tax-guru"
itg years

As a library in your project:

uv add "india-tax-guru @ git+https://github.com/anivar/india-tax-guru"
# or: pip install "india-tax-guru @ git+https://github.com/anivar/india-tax-guru"

The library has zero runtime dependencies; click is pulled in only by the [cli] extra.

From a clone (development):

uv sync --group dev
uv run itg years

Usage

itg compare path/to/profile.json     # old vs new regime, full breakdown
itg advise path/to/profile.json      # what to change and what it's worth
itg optimize-ctc path/to/ctc_input.json

(Prefix with uv run when working from a clone.)

See docs/profile_schema.md for the input JSON shape, or use the library directly:

from india_tax_guru.models import TaxpayerProfile, AgeBand
from india_tax_guru.regime import compare_regimes
from india_tax_guru.rules import get_rules

profile = TaxpayerProfile(assessment_year="2026-27", age_band=AgeBand.BELOW_60, ...)
rules = get_rules(profile.assessment_year)
comparison = compare_regimes(profile, rules)
print(comparison.recommended, comparison.savings)

Development

uv sync --group dev
uv run pytest
uv run ruff check .

Disclaimer

This tool encodes the author's best understanding of Indian income-tax rules at the time each rules/ay*.py module was written. Tax law is complex and changes frequently; verify figures against CBDT circulars / a qualified CA before relying on this for an actual filing. No warranty, express or implied — see LICENSE.

License

MIT — see LICENSE.