Zodiacally Ephemeris
Our own engine. Your next idea.
The astronomical engine behind Zodiacally — a JavaScript library for astrology apps, developers and agents. Built on NASA/JPL’s DE440 planetary data, with no native add-on and no Swiss Ephemeris code. Source, tests and a benchmark anyone can rerun are on GitHub.
In Node.js our engine is 14 to 24 times faster per chart than the Swiss Ephemeris via sweph once both are loaded. The Swiss Ephemeris starts faster: loading the library and computing the first chart in a fresh process (Node.js start-up excluded) takes 19.2 ms with our engine and 10.4 ms with the Swiss Ephemeris.
Version 1.1.1 · Node.js · Local, with no API key
Built by us
Built on JPL data. Checkable by anyone.
Used by Zodiacally since October 2026. The engine evaluates precomputed, checksummed data tables that a published build script generates from the NASA/JPL data. A rebuild on macOS (Apple silicon) reproduced every table byte for byte, and on Linux numerically equivalently (within 0.00000002″); a monthly CI job checks this. At runtime it needs no network connection, native add-on or Swiss Ephemeris code. Astronomical calculation and astrological interpretation remain separate steps.
12Bodies and points: Sun through Pluto, Chiron and the true lunar node
1899–2101Data coverage from 1899-01-01 00:00:04 to 2101-01-13 23:58:50 UTC; exact bounds in info().coverage
JavaScriptLibrary for CommonJS and ES modules with TypeScript types, plus a command line for agents
Tropical apparent geocentric longitudes and their speeds, without latitude or distance; Placidus and whole-sign houses. A focused feature set: the Swiss Ephemeris covers more dates, objects and calculation modes.
Measured on 8 November 2026
14 to 24 times faster per chart, once loaded.
Both libraries received the same 1,024 different inputs through their Node.js APIs, in a single thread. After warm-up, seven rounds ran in alternating order; the figures are medians.
| Workload | Zodiacally Ephemeris | Swiss Ephemeris (sweph) | Factor |
|---|
| Full chart: 12 positions with speeds + Placidus houses | 5.50 µs | 77.05 µs | 14.0× |
|---|
| 12 positions with speeds at one date (no houses) | 1.51 µs | 69.12 µs | 45.7× |
|---|
| Placidus houses only | 3.57 µs | 9.51 µs | 2.7× |
|---|
| One position with speed, new date every call | 0.121 µs | 17.32 µs | 143× |
|---|
Per body, the second row works out at 0.126 µs versus 5.760 µs. The last row is the most favourable case for our engine: with a new date on every call, the Swiss Ephemeris cannot reuse any work per date. The chart rows are the representative ones.
Apple M1 · macOS · Node 25.8.2 · sweph 2.10.3-7 / Swiss Ephemeris 2.10.03. On a second machine (AMD Ryzen 5 5600G, Windows, Node 22.14.0, 4 November 2026, measured with the earlier single-workload version of the script, on different inputs from the same ranges): 7.54 µs versus 177.53 µs per chart (23.5×).
The Swiss Ephemeris starts faster: loading the library and computing the first chart in a fresh process (Node.js start-up excluded) takes 19.2 ms with our engine and 10.4 ms with the Swiss Ephemeris. The extra time arises because our engine loads Node’s crypto module and reads and verifies the SHA-256 of its 8 MB of tables on first use. Counting start-up, ours is ahead in total time only after about 124 charts in one process. The comparison is also not symmetric: our engine evaluates precomputed apparent longitudes, while the Swiss Ephemeris computes positions from its data at run time, including latitude, distance and their speeds. It is called through the sweph native binding; a C program calling the Swiss Ephemeris directly was not measured.
Agreement: Swiss Ephemeris and NASA/JPL Horizons
- Swiss Ephemeris with its default files, 20,000 instants 1900–2100, the same instant for both: in this run, planets more than 1° from the Sun differ by 1.6–4.4 mas (p99; at most 9.0 mas), the Sun by at most 3.0 mas and the Moon by at most 8.7 mas (1 mas = 0.001″). Larger differences have known causes: for a planet next to or behind the solar disc the libraries model light deflection differently (up to 0.27″ in this run, 0.55″ in the October internal report); the true node differs by up to 0.18″ (p99 0.025″; 0.26″ in the October internal report), and Chiron by up to 0.93″, because the two use different orbit solutions.
- Zodiac signs: at the same instant the sign differed in 0 of 240,000 positions, and the displayed whole arcminute in 61 (mostly Chiron). That only rules out gross errors. End to end from a UTC reading (internal report, 55,205 instants 1899–2101, each library with its own time model; dates and distinct-instant counts from a November 2026 rerun of the same internal harness and seed, see docs/accuracy.md), 4 body positions fell in a different sign: three from November 2049 on (Mercury 2049-11-11, Moon 2054-03-27 and 2066-04-21), where the two libraries map a future UTC to TT differently, and one Jupiter position on 2021-05-13 that lay within 0.001″ of the 330° boundary. The Ascendant fell in a different sign at 41 instants, all from 2044 on (the raw report counts 77 because it counts the Placidus and whole-sign runs separately).
- Placidus houses at the same UT1: at latitudes up to 66°, from 1900 to 2049, the Ascendant, MC and cusps differ by at most 0.027″ (p99 0.003″ for the Ascendant, 0.001″ for the MC, 0.002″ for all cusps). Right next to the polar circle (|latitude| ≈ 66.5°) the Ascendant is singular: up to 0.67″ (October internal report; 0.66″ for Placidus). From 2050 the MC differs by an almost constant 1.87″, because the Swiss Ephemeris changes its sidereal-time model there. The Ascendant and cusps amplify that step: up to 30″ at latitudes up to 66° (largest above 60°) in the 8 November run, and up to about 20′ (1,188″) right next to the polar circle in the October internal report. That can change the displayed arcminute of a cusp.
- Swiss Ephemeris reading JPL’s DE440 file: with the same JPL data on both sides, the two codes differ by 1.5–5.7 mas (p99). This shows that both compute the apparent place consistently. It does not rank them: our engine is built from DE440 itself, so it is naturally close to a DE440 calculation.
- NASA/JPL Horizons, 200 instants: for the Sun to Pluto both libraries differ from Horizons by up to 0.245″ and from each other by at most 0.0073″. The shared part arises because JPL gives this Horizons quantity in the IAU 1976/1980 frame, while both libraries use the IAU 2006/2000 models, so Horizons cannot rank the two. Horizons itself currently computes from DE441 (shown in its output header), so this comparison does not favour the DE440 build either. Chiron follows JPL’s current orbit solution #171 (the one JPL Horizons serves); the Swiss Ephemeris asteroid file uses a different solution and differs from it by up to about 0.69″. Because the Chiron table was built from that same solution, this is a consistency check, not an independent accuracy result.
Apart from Chiron, the true node, planets right next to the Sun, the Ascendant right next to the polar circle and the houses from 2050, the differences are a few milliarcseconds for the bodies and at most 0.027″ for the houses, thousands of times smaller than the arcminute a chart displays. End to end from a UTC reading, the two time models also matter: in the October internal report the MC differs by up to about 1″ from 1972 to 2022; for future dates positions differ by up to 3.4″ (Moon) by 2049 and 15.6″ by 2100, and houses by much more, because every ephemeris must predict the Earth’s rotation. We do not claim that either library is more accurate.
Figures from the “October internal report” and the “internal report” come from Zodiacally’s internal comparison harness (documented in docs/accuracy.md); unlike the 8 November benchmark, that harness is not part of the public repository. On provenance: no Swiss Ephemeris code was read or used, and no coefficient was fitted to its output; the tables come only from JPL data. We did, however, choose three modelling conventions by comparing residuals with Swiss Ephemeris output: light deflection by the Sun only (as JPL Horizons also uses), the model for light passing behind the solar disc, and the definition of the true node. Part of the agreement is therefore by construction, which is why the Horizons comparison is reported too.
Run the comparison yourself
Needs Node.js 22.12 or later. The benchmark scripts are MIT-licensed; npm install there adds sweph (AGPL-3.0/LGPL-3.0) for the comparison only, and the script downloads the official Swiss Ephemeris files with checksums. Their licence terms also apply to the comparison. sweph ships prebuilt binaries for macOS arm64, Linux x64 and arm64, and Windows x64; on other platforms npm install there needs a C toolchain. Add --jpl to fetch-swiss-data.mjs and agreement.mjs for the DE440 comparison (102 MB download).
git clone https://github.com/Kululenko/zodiacally-ephemeris.git
cd zodiacally-ephemeris/benchmark && npm install
node fetch-swiss-data.mjs && node speed.mjs && node agreement.mjs && node horizons.mjs && node summary.mjs
Comparison
An alternative to the Swiss Ephemeris?
The Swiss Ephemeris is the long-established standard in astrology software. For the standard chart in JavaScript — Sun to Pluto, Chiron, the true node, Placidus or whole sign, 1899 to 2101 — the position differences at the same instant lie far below what a chart displays (exceptions such as the houses from 2050 are listed above); once loaded, our engine is faster per chart (it starts more slowly) and needs no native add-on. If you need more, such as sidereal zodiacs, asteroids, fixed stars, other house systems or dates outside that range, stay with the Swiss Ephemeris.
| Zodiacally Ephemeris 1.1.1 | Swiss Ephemeris 2.10 |
|---|
| Planetary data | NASA/JPL DE440, recommended by JPL for modern dates; Chiron from JPL orbit solution #171 | SE1 files of 2026 built from NASA/JPL DE441, the long-span companion of DE440; the two differ by a few metres at most |
|---|
| Full chart in Node.js (loaded) | 5.5 µs | 77.0 µs |
|---|
| Load + first chart in a new process | 19.2 ms | 10.4 ms (faster) |
|---|
| Returned per call | apparent longitude and its speed | longitude, latitude, distance and their speeds |
|---|
| Installation | pure JavaScript, zero dependencies, data included; from GitHub (npm release pending) | in Node.js: native add-on (sweph) plus separately deployed data files |
|---|
| Scope | 12 bodies and points, Placidus and whole sign, 1899–2101 | far more bodies, house systems and calculation modes (including sidereal), 13,201 BC to AD 17,191 |
|---|
| Licence | source-available, not open source; free for any use, including commercial, with a visible attribution line in every reading or result shown or delivered to others; white-label without attribution: one-time €690 plus VAT where applicable | AGPL-3.0 or the Swiss Ephemeris Professional License |
|---|
The Swiss Ephemeris is developed by Astrodienst AG, which has not reviewed or endorsed this comparison. Detailed comparison and migration guide
For your project
Free to use. License it without attribution.
You can use the engine free of charge for anything, including commercial work. The main condition: every reading or result you show or deliver to others carries the attribution line; results you only use yourself need none. If you would rather not show the line, a white-label licence is available on request for a one-time fee of €690 plus VAT where applicable.
Free — with attribution
€0
For any use, including commercial
- For your own apps and websites, client work, paid readings, agents and APIs.
- With a visible attribution line in every reading or result you show or deliver to others; for short formats, a clearly labelled credit reachable from the output is enough.
- You may not distribute the engine itself as a standalone library, whether free or paid.
- No registration, separate agreement or licence fee.
White-label
€690 plus VAT where applicable
One-time perpetual licence, on request
- No attribution line in your readings and results.
- For you or your company: all your own products and services, and client projects you build.
- Covers the version current at purchase and every release published within 12 months, perpetually and with no recurring fee.
- Source and rights notices remain in the software package.
The source code is available and free to use, also commercially, but the licence is not an OSI open-source licence. You may distribute the engine as part of your own application. As a standalone library, package, port or data product, however, it may not be published, distributed, sold, rented or sublicensed, whether for a fee or free of charge; forks for proposing changes are allowed. If you do not meet the attribution requirement or these distribution rules, your rights end; they are restored if you remedy the breach within 30 days. The white-label licence covers one person or company with all its own products and client projects, for the version current at purchase and every release published within 12 months, perpetually and with no recurring licence fee. It does not promise hosting, support, maintenance or new releases. We grant white-label licences individually on request and confirm the terms and invoice before purchase; downloading the package alone grants no white-label licence.
Read the licence terms · 7.5 MB · Version and SHA-256
Developers and agents
Pass on the calculation and its provenance.
The package includes a library for CommonJS and ES modules with TypeScript types, and a CLI with JSON input and output. The agent interface returns calculated values, version information and the required attribution line. Unless you hold a white-label licence, that line must be visible in every completed reading or result you show or deliver to another person — including PDFs, emails and chat responses. For short formats such as push notifications, widgets or voice output, a clearly labelled credit reachable from the output is enough. If you pass results on through an API, include the line in the response and require your recipients to display it.
Astronomical calculations: Zodiacally Ephemeris, based on JPL DE440 (NASA/JPL). zodiacally.com/ephemeris
Example of the visible credit line in a readingAn internal instruction to the agent, a system prompt or a hidden metadata field is not enough. Only a white-label licence waives the attribution. The library does not verify purchases or enforce rendering by third-party agents; correct output is a licence and integration requirement.
npm install github:Kululenko/zodiacally-ephemeris
npx --no-install zodiacally-ephemeris --info
The npm release is pending; until then, install directly from GitHub. npx --no-install only runs the copy you have just installed; alternatively: node node_modules/zodiacally-ephemeris/cli.cjs --info. The CLI and the agent interface do not accept a leap second (23:59:60); call utcToJd() directly for that.
FAQ
Frequently asked questions
Is it an alternative to the Swiss Ephemeris?
For the standard astrological chart in Node.js, yes: Sun to Pluto, Chiron and the true lunar node, Placidus or whole-sign houses, 1899 to 2101. Compared at the same instant, the zodiac sign was the same in all 240,000 positions of our benchmark, which only rules out gross errors. End to end from a UTC reading (internal report, 55,205 instants 1899–2101, each library with its own time model), 4 body positions fell in a different sign: three from November 2049 on (Mercury 2049-11-11, Moon 2054-03-27 and 2066-04-21), where the two libraries map a future UTC to TT differently, and one Jupiter position on 2021-05-13 that lay within 0.001″ of the 330° boundary. The Ascendant fell in a different sign at 41 instants, all from 2044 on (the raw report counts 77 because it counts the Placidus and whole-sign runs separately). It offers no sidereal zodiacs, other asteroids, fixed stars, further house systems or dates outside that range; for those, use the Swiss Ephemeris.
How accurate is it?
It is built on NASA/JPL’s DE440, the JPL ephemeris recommended for modern dates. The Swiss Ephemeris’ current files are built from the long-span companion DE441; the two differ by a few metres at most, which is irrelevant for astrology. At the same instant the two libraries agree to a few milliarcseconds for the Sun, the Moon and planets more than 1° from the Sun, and to at most 0.027″ for Placidus houses before 2050 at latitudes up to 66°, thousands of times smaller than the arcminute a chart displays. Three modelling conventions (light deflection by the Sun only, the model for light passing behind the solar disc and the definition of the true node) were chosen by comparing residuals with Swiss Ephemeris output, so part of this agreement is by construction. Larger differences have known causes: light deflection right next to the Sun (up to 0.55″) and the true node (up to 0.26″), both observed in the October internal report; Chiron (up to 0.93″, different orbit solutions); the Ascendant right next to the polar circle (up to 0.67″ in the October internal report; 0.66″ for Placidus); and the houses from 2050, where the Swiss Ephemeris changes its sidereal-time model: the MC then differs by an almost constant 1.87″, and the Ascendant and cusps by up to 30″ at latitudes up to 66° in the 8 November run and by up to about 20′ right next to the polar circle in the October internal report. End to end from a UTC reading, the two time models also matter: in the October internal report the MC differs by up to about 1″ from 1972 to 2022; for future dates positions differ by up to 3.4″ (Moon) by 2049 and 15.6″ by 2100, and houses by much more, because every ephemeris must predict the Earth’s rotation. A comparison with NASA/JPL Horizons does not rank the two either, because a shared frame term dominates there; we do not claim that either library is more accurate.
How much faster is it?
In Node.js, once both libraries were loaded, a full chart with 12 positions and Placidus houses took 5.5 µs instead of 77.0 µs on an Apple M1 (14.0 times faster) and 7.5 µs instead of 177.5 µs on an AMD Ryzen 5 under Windows (23.5 times, measured with the earlier single-workload version of the benchmark script, on different inputs from the same ranges). The Swiss Ephemeris starts faster: loading the library and computing the first chart in a fresh process (Node.js start-up excluded) took 19.2 ms with our engine and 10.4 ms with the Swiss Ephemeris. The extra time arises because our engine loads Node’s crypto module and reads and verifies the SHA-256 of its 8 MB of tables on first use. The comparison is not symmetric: our engine evaluates precomputed apparent longitudes, while the Swiss Ephemeris computes positions from its data at run time and also returns latitude, distance and their speeds. The Swiss Ephemeris was called through the sweph native binding; a C program calling it directly was not measured.
Is it free?
Yes: free for any use, including commercial, as long as every reading shown to others carries the attribution line; white-label licence without attribution: a one-time fee of €690 plus VAT where applicable. The line must be visible in every reading or result you show or deliver to another person, whether or not they pay, including PDFs, emails and chat responses, in the language of the reading. In English it reads “Astronomical calculations: Zodiacally Ephemeris, based on JPL DE440 (NASA/JPL). https://www.zodiacally.com/ephemeris/”. Results you only use yourself need no line. For short formats such as push notifications, widgets, social media posts, calendar entries or voice output, a clearly labelled credit reachable from the output, such as a linked caption, is sufficient. If you pass results on through an API, include the line in the response and require your recipients to display it. The white-label licence covers you or your company with all your own products and client projects, for the version current at purchase and every release published within 12 months, perpetually; we grant it individually on request. The source code is published on GitHub, but it is not open source: the engine may not be published, distributed, sold, rented or sublicensed as a standalone library, whether for a fee or free of charge; forks for proposing changes are allowed.
Can I check the results myself?
Yes. The GitHub repository contains the benchmark scripts (MIT-licensed), the raw NASA/JPL Horizons responses, the convention study scripts and the build that regenerates the data tables from NASA/JPL’s DE440s file and JPL Horizons data for Chiron (orbit solution #171). A rebuild on macOS (Apple silicon) reproduced every table byte for byte, and on Linux numerically equivalently (within 0.00000002″); a monthly CI job checks this. docs/verify.md lists five steps. Figures labelled “October internal report” or “internal report” come from our internal comparison harness, which is not in the repository.
Does it run in the browser?
No. It runs in Node.js 22.12 or later with file access, because it reads its 8 MB of data tables from disk.
Planetary data source: NASA/JPL DE440. Source attribution does not imply NASA/JPL endorsement of astrological interpretations.