Vedākṣha — Vision from Vedas

Clean-room Rust ephemeris and Vedic astrology engine, built for the agentic-AI era. Sub-arcsecond planetary precision, every algorithm traced to a primary source, every chart emitted as a queryable property graph.

Celestial computation. Agentic precision.

Website · Docs · Playground · API reference · Blog

clean-room · sub-arcsecond vs JPL Horizons · 820 tests + 24,350 oracle rows · MCP-native · BSL 1.1 → Apache 2.0

Quick start

use vedaksha::prelude::*;

let jd = calendar_to_jd(2024, 3, 20, 12.0);
let chart = compute_chart(jd, 28.6139, 77.2090, &ChartConfig::vedic());

cargo add vedaksha          # Rust
pip install vedaksha        # Python (runs the engine via WebAssembly)
npm install vedaksha-wasm   # WebAssembly / JavaScript

Compute janam kundali (natal charts), panchanga (the five limbs of the day), vimshottari and other dashas, nakshatras, vargas (divisional charts), shadbala, ashtakavarga, muhurta, and transits/gochara — from a sub-arcsecond ephemeris (VSOP87A, ELP/MPP02, JPL DE440s/DE441), in Rust, Python, WebAssembly, or as an MCP server for AI agents.

Related MCP server: Precision astronomical ephemeris and planetary positions via the Swiss Ephemeris.

Why Vedākṣha

• Clean-room, cited. Every module that implements a cited algorithm carries a // Source: doc-comment pointing at the primary paper or treatise (VSOP87A, ELP/MPP02, IAU standards, BPHS, Jaimini) — never derived from other software, no GPL contamination. See DATA_PROVENANCE.md and docs/audit/.

• Sub-arcsecond, measured. 820 tests on every push (Ubuntu and macOS); a scheduled full run adds 24,350 oracle comparisons against JPL Horizons / DE441 — mean residual 0.106″ over 1900–2025. Every number in Accuracy is printed by a test you can run.

• Agentic-AI-native. A 15-tool Model Context Protocol server, and every chart is a property graph you can query in Cypher, SurrealQL, or JSON-LD.

• Runs everywhere. One Rust codebase → native, Python (pip install vedaksha, the engine hosted via WebAssembly — no Rust toolchain, one py3-none-any wheel), in-browser WebAssembly (no data files), and a multi-arch containerized MCP server. No FFI to a C library, no platform-specific build.

• Jyotish in the type system. Nakshatras, dashas, vargas, shadbala, ayanamshas — first-class, not a Western afterthought.

In production

Vedākṣha is the calculation engine under ArthIQ Labs' Jyotish properties:

Workspace

Two ephemeris providers

The AnalyticalProvider evaluates VSOP87A (Bretagnon & Francou 1988) for planets and ELP/MPP02 (Chapront 2002) for the Moon — all coefficients are compile-time constants, so there are no runtime data files.

Computation pipeline

JPL DE440 SPK → Chebyshev evaluation → ICRS barycentric
  → light-time correction → precession (IAU 2006 P03, 5th-order)
  → nutation (IAU 2000B) → frame bias (ICRS→J2000)
  → aberration → ecliptic coordinates

Zero-data path (WASM / edge):

VSOP87A / ELP coefficients (compiled) → Poisson series evaluation
  → heliocentric ecliptic → equatorial rotation → barycentric ICRS
  → same downstream pipeline

Delta T: IERS measured table (1620–2025) + Espenak–Meeus predictions to 2050.

Vedic astrology

First-class Jyotish, drawn from primary classical sources.

• Nakshatras — 27 lunar mansions with padas, lords, symbols, deities

• Dashas — Vimshottari (120-yr), Yogini (36-yr), Ashtottari (108-yr), and Chara & Narayana (Jaimini, sign-based)

• Vargas — all 16 divisional charts (D-1 Rashi → D-60 Shashtiamsha)

• Shadbala — complete six-component planetary strength, with Ishta / Kashta phala

• Ayanamsha — 44 sidereal systems (Lahiri, Raman, KP, Fagan-Bradley, +40)

• Lunar nodes — Mean, True (Meeus 5-term, ~0.09°), and Osculating (0.6″ max vs JPL DE441 over 1900–2100) — KP sub-lord ready

• Panchanga — full five limbs: Tithi (paksha, lord), Vara (Rahu / Gulika Kalam), Nakshatra (deity, yoni, nadi), Yoga (27), Karana (60)

• Drishti — graded aspects: Full, ¾ (75%), ½ (50%), ¼ (25%) per BPHS Ch. 26

AI-native: MCP + property graph

Every computation produces a property graph, not flat structs — so an agent can ask "which planets aspect the 7th-house lord?" as a graph query instead of re-implementing chart logic. The MCP server exposes 15 tools, discoverable with a single tools/list call:

compute_natal_chart · compute_dasha · compute_vargas · compute_karakas · compute_combustion · compute_shadbala · compute_ashtakavarga · compute_transit · compute_gochara · search_transits · search_muhurta · compute_panchanga · compute_drishti · compute_bhavas · emit_graph

cargo install vedaksha-mcp
vedaksha-mcp                                          # stdio (Claude Desktop, Cursor, VS Code)
VEDAKSHA_MCP_TOKEN=… vedaksha-mcp --http --port 3100  # HTTP transport (auth required)
docker run -e VEDAKSHA_MCP_TOKEN=… -p 3100:3100 ghcr.io/arthiqlabs/vedaksha-mcp

HTTP mode requires a bearer token (Authorization: Bearer <token> on every POST); the server refuses to start without VEDAKSHA_MCP_TOKEN unless you pass --insecure-no-auth for a trusted-network deployment. /health and the informational GET stay open. The Docker image is multi-arch (amd64 + arm64).

The tool surface is generated from the Rust definitions and locked by a snapshot test, so the published catalog can't silently drift from the code.

Accuracy

Every figure below is printed by a named test. Reproduce them with bash scripts/download_de440s.sh then cargo test -p vedaksha-ephem-core --release -- --include-ignored --nocapture.

SpkReader vs JPL Horizons (DE441) — oracle_comparison.rs, over the 24,350 rows in tests/oracle_jpl/ (10 bodies × 2,435 dates, 1900–2100). Horizons serves DE441, so this measures our DE440s pipeline against an independent kernel:

15,349 of 15,350 comparisons before 2026 are sub-arcsecond. Past 2025 the residual is dominated by ΔT prediction, not ephemeris error: our Espenak & Meeus extrapolation and Horizons' own ΔT diverge by ~68 s at 2099, which shows up in proportion to a body's angular rate (the Moon, at 0.64″/s, picks up ~45″; Pluto, essentially none). At 2099-02-06 the Sun, Moon, Mercury, Venus and Mars — rates spanning 0.03–0.64″/s — all imply the same 66–71 s offset. ΔT beyond the IERS measured record is unpredictable in principle, not a defect we can fix.

AnalyticalProvider vs JPL Horizons (DE441) — analytical_oracle.rs, the same fixture over 1900–2025 (13,815 comparisons). VSOP87A is a truncated analytical theory, so it is necessarily looser than the numerical kernel:

Overall mean 2.06″. analytical_accuracy.rs reports a friendlier 13.09″ max for the same provider, but it samples 10 dates against this test's 2,435 per body — the sparse grid never lands near Venus's worst case. The table above is the better-sampled number and the one to trust.

ELP/MPP02 Moon vs JPL Horizons (DE441) — lunar_horizons.rs, live-fetched over −3000…+3000 CE: 0.015″ at J2000 (tolerance 0.06″), 0.020–0.053″ across 1500–2500 CE, degrading to 85.5″ in deep antiquity where ELP/MPP02's own published precision is the limit.

Ayanamsha — Lahiri, Fagan-Bradley and KP are checked at J2000 to 0.003–0.005° (sidereal.rs), propagating documented epoch anchors with IAU 1976 / Newcomb precession. Every anchor's origin is listed in DATA_PROVENANCE.md. The other 41 systems are range-checked only — no accuracy figure is claimed for them.

Dasha totals and nakshatra boundaries are covered by invariant tests (vimshottari.rs, nakshatra.rs). Those verify internal consistency — that our BPHS constants sum to 120 years, that boundaries tile the circle — and are not comparisons against an external reference. House-cusp accuracy is not measured against any reference.

Install

Published: crates.io — 7 crates (vedaksha, vedaksha-math, vedaksha-ephem-core, vedaksha-astro, vedaksha-vedic, vedaksha-graph, vedaksha-mcp) · PyPI vedaksha · npm vedaksha-wasm · Docker ghcr.io/arthiqlabs/vedaksha-mcp (multi-arch).

See bindings/python/ for the Python package (library, vedaksha CLI, self-hostable MCP server, and optional FastAPI REST).

License

Business Source License 1.1.

• Non-commercial use — free (personal projects, research, education, internal tools).

• Commercial use — $500 one-time per organization. Purchase →

• Converts to Apache 2.0 five years after each version's release date.

See LICENSE for full terms.

Copyright © 2026 ArthIQ Labs LLC · Licensed under BSL 1.1.