AstroMansion

Official Python client for the AstroMansion astrology API.

Nothing is computed locally. Every call reaches https://api.astromansion.com, which owns the ephemeris, your plan, your quota and your rate limit. The package never opens a feature the server did not grant.

Install

pip install astromansion

Python 3.10 or newer. The only dependency is httpx, installed with its SOCKS support so the client still works behind a corporate proxy, a local SOCKS proxy or Tor without anything further to install.

Get an API key

Create an account at astromansion.com, open your account page and generate a key. Requests made with it count against that account, under the plan it already has.

First chart

Put the key in the environment rather than in your source:

export ASTROMANSION_API_KEY="your key"

from astromansion import AstroMansion

client = AstroMansion()

chart = client.natal(
    date="1990-07-19",
    time="14:30",
    lat=41.0082,
    lon=28.9784,
    timezone=3,
)

print(chart.summary.Sun.sign)  # Cancer
print(chart.planets[0].house)  # 9

Birth data is passed flat. The API nests it under birth; the client does that for you.

Fields: date as YYYY-MM-DD, time as HH:MM (omit if unknown), lat and lon in decimal degrees, timezone as an hour offset or an IANA zone name, houses for a house system.

You can pass a mapping instead of keywords, but not both at once:

chart = client.natal({"date": "1990-07-19", "lat": 41.0082, "lon": 28.9784})

Where the key comes from

In order: the api_key argument, then astromansion.set_api_key(...), then ASTROMANSION_API_KEY. With none of them the constructor raises AuthenticationError, before a connection is opened and before the proxy environment is read, so a forgotten key is reported as a forgotten key.

client = AstroMansion(api_key="your key")

astromansion.reset() drops the module key and the shared client, which is what a test wants between cases.

Quick use

For a notebook or a one-file script:

import astromansion as am

am.set_api_key("your key")  # or rely on the environment
chart = am.natal(date="1990-07-19", lat=41.0082, lon=28.9784)

Applications should build a client instead: it holds a connection pool, and two of them can carry two different keys.

Async

from astromansion import AsyncAstroMansion

async with AsyncAstroMansion() as client:
    chart = await client.natal(
        date="1990-07-19",
        time="14:30",
        lat=41.0082,
        lon=28.9784,
        timezone=3,
    )

Same method names, same arguments, same exceptions. Python cannot make one class serve both, so the bare name is synchronous and Async marks the other, as in httpx, openai and anthropic.

Reading a response

The response is the server's own JSON, readable either way:

chart.summary.Sun.sign
chart["summary"]["Sun"]["sign"]
chart.to_dict()

Nothing is remodelled, so a field the API adds reaches you instead of being dropped, and no field it did not send is invented.

Some endpoints answer with the data itself and some wrap it as {"technique": ..., "result": ...}. The client opens that wrapper, so every response reads the same way and you never have to remember which kind you called:

chart = client.vedic_chart(date="1990-07-19", lat=41.0082, lon=28.9784)

chart.data          # the payload, wrapper removed
chart.technique     # what the server called it, when it said
chart.raw           # the untouched body, wrapper included

A body that carries a real result field of its own is left alone; the wrapper is recognised by its exact shape rather than by the presence of a name that data is allowed to use.

Arabic lots, fixed stars and the rest of the catalog

natal answers with the chart proper. Anything beyond it is named on chart, through options.categories:

lots = client.chart(
    date="1990-07-19", time="14:30",
    lat=41.0082, lon=28.9784, timezone=3,
    options={"categories": ["arabic_lots"]},
)

for lot in lots.data["bodies"]["arabic_lots"]:
    print(lot["name"], lot["sign"], lot["dms"], lot["house"])

Bodies come back grouped under the category that produced them, so read the group you asked for. The catalog publishes 38 Arabic lots and 908 named fixed-star entries, along with planets, dwarfs, asteroids, centaurs, comets, hypotheticals, points, advanced_points, lilith, planetary_nodes, exoplanets, antiscia and eclipses. The API still accepts the retired moons selector as an input alias, but responses use antiscia.

The fixed_stars calculation category resolves 906 unambiguous targets. Isis and Ketu remain discoverable catalog names, but a bare-name lookup for either resolves the higher-priority non-stellar point with the same name.

chart answers one page at a time, and a large category is more than one page. Use bodies to read the whole of it and let the client follow the paging:

found = client.bodies(
    "fixed_stars", "arabic_lots",
    date="1990-07-19", time="14:30",
    lat=41.0082, lon=28.9784, timezone=3,
)

len(found["fixed_stars"])  # 906, in six requests
len(found["arabic_lots"])  # 38

Naming several categories reads them in one walk. The return is always a mapping keyed by the categories you asked for, one or several, so the shape never depends on how many.

Each page costs a network round trip and the calculation inside it is a rounding error beside that, so bodies asks for the largest page the API serves. page_size lowers it, and lowering it buys nothing: the 906 calculation targets take six requests at the default page size.

Some categories are enormous. asteroids holds twenty-eight thousand bodies, which is a hundred and seventy-seven round trips with nothing printed until the last one lands, and a program that prints nothing for a minute cannot be told apart from one that has hung. A walk that large stops after the first page and tells you the size:

am.bodies("asteroids", date="2000-01-01", lat=51.4779, lon=0.0)
# ValidationError: asteroids holds 28214 bodies, which is 177 requests and
# will print nothing until the last one lands. Pass confirm_large=True ...

Read it anyway, or watch it arrive:

found = client.bodies(
    "asteroids", date="2000-01-01", lat=51.4779, lon=0.0,
    confirm_large=True,
    on_page=lambda done, left, rows: print(f"{done} done, {left} to go"),
)

For a handful of named bodies there is no walk at all:

page = client.chart(
    date="2000-01-01", lat=51.4779, lon=0.0,
    options={"include": ["Ceres", "Pallas", "Juno", "Vesta"],
             "exclusive": True},
)

for rows in page.data["bodies"].values():
    for body in rows:
        print(body["name"], body["sign"], body["dms"])

Read every group rather than the one you expected. Bodies are filed under the category the catalog puts them in, not the category you were thinking of when you named them: Ceres was an asteroid in 1801 and has been a dwarf planet since 2006, so it arrives under dwarfs while Pallas, Juno and Vesta arrive under asteroids. A loop over bodies["asteroids"] alone finds three of the four and reports the fourth as missing.

exclusive narrows the catalog, not the chart. The angles and the derived points are calculated either way, so points and advanced_points are still in the answer. Filter by name if you want only what you asked for:

wanted = {"Ceres", "Pallas", "Juno", "Vesta"}

for rows in page.data["bodies"].values():
    for body in rows:
        if body["name"] in wanted:
            print(body["name"], body["sign"], body["dms"])

Reach for chart directly when you want one page rather than the category, and read the group with .get. The chart's own bodies are calculated alongside the categories you name and the page is a window over that whole selection, so a small catalog_limit can fill the first page with planets and angles before a single star appears, leaving no fixed_stars key at all. catalog_page.total counts the same way, the whole selection rather than the category.

There is no ceiling on how many you may read this way. options.all_bodies walks every category at once and needs the full-catalog scope that comes with Pro and Enterprise; naming the categories yourself does not.

MansionSQL

A chart is a table of bodies, a table of houses and a table of aspects, and the questions people ask of one are the questions SQL was written for: which planets are retrograde, which aspects are inside a degree, how the signs are distributed. query sends a SELECT and the server calculates the chart and answers it.

rows = client.query(
    birth={"date": "2000-01-01", "time": "12:00",
           "lat": 51.4779, "lon": 0.0, "timezone": 0},
    sql="SELECT name, sign, dms FROM planets WHERE retrograde = 1",
)

for row in rows.data:
    print(row["name"], row["sign"], row["dms"])
# Saturn Taurus 10°23′44″

query takes the request body rather than flat birth fields, because a statement may name two charts and a moment, and each of them needs its own birth block. It needs the advanced scope.

Tables

Every table but three carries the body columns: code, name, longitude, latitude, distance, speed, sign, sign_index, house, retrograde, dms, category.

Three have columns of their own:

natal_b needs partner, and a statement naming it without one is refused. transits uses moment, and without one it is the current instant at the birth location.

There is no asteroids table. One chart holds at most 192 bodies and the category is twenty-eight thousand, so the query is refused rather than silently reduced to the first 192:

client.query(birth={...}, sql="SELECT name FROM asteroids")
# ValidationError: The query asks for more bodies than one chart can hold.
# Narrow it with WHERE name = 'Pallas', a minor planet number such as
# WHERE name = '1198', or WHERE name IN (...). Use /v1/catalog to look a
# body up by name.

stars answers 953 rows, the fixed_stars catalog lists 908 names, and the category walk returns 906 unambiguous calculation targets. These are different selections, so a count taken from one does not describe the others. The table contains every catalog row the engine places against a stellar position: 908 fixed-star rows, 40 exoplanets and 5 galactic and deep-sky points filed under advanced_points. One row per published name means counting rows and names gives the same answer.

lots and the arabic_lots category do not divide that way. Both answer 38: the table gives the lots computed for one chart, the category gives the names those lots are looked up by, and they are the same 38 lots.

SELECT only. Anything else is refused before it reaches the engine, so a statement cannot write, drop or reach outside the chart it was given.

client.query(birth={...}, sql="DELETE FROM planets")
# ValidationError: expected SELECT at offset 0

A hundred rows

An answer is capped at a hundred rows, and a statement that would exceed it is refused rather than truncated, so a partial answer never arrives looking whole:

client.query(birth={...}, sql="SELECT name FROM stars")
# ValidationError: MansionSQL returned 100 rows of 953; add LIMIT or narrow
# the query

Add LIMIT, or narrow with WHERE. COUNT(*) is one row, so counting a large table is always available.

What the dialect supports

WHERE, GROUP BY, HAVING, ORDER BY, LIMIT, OFFSET, DISTINCT, JOIN, AS, IN, NOT IN, LIKE, BETWEEN, IS NULL, EXISTS, CASE WHEN, and comparison and arithmetic on any column.

Clauses apply in the SQL order, which is worth stating because it decides what HAVING sees and what the hundred-row cap counts:

FROM -> WHERE -> GROUP BY -> HAVING -> SELECT -> ORDER BY -> LIMIT

# The tightest aspects first.
"SELECT a, b, aspect, orb FROM aspects WHERE orb < 1 ORDER BY orb LIMIT 20"

# How the chart is distributed.
"SELECT sign, COUNT(*) AS n FROM planets GROUP BY sign ORDER BY n DESC"

# Angles only.
"SELECT number, sign, dms FROM houses WHERE number IN (1, 10)"

# Each planet with the sign on the cusp of the house it occupies.
"SELECT p.name, h.sign FROM planets p JOIN houses h ON p.house = h.number"

# The brightest fixed stars.
"SELECT name, sign, dms, magnitude FROM stars ORDER BY magnitude LIMIT 10"

Aggregates

AVG and SUM refuse a column that is not numeric rather than skipping the rows they cannot read. Skipping is the dangerous answer: over a column that is numeric for some rows and text for others it returns the average of part of the column with nothing to say the rest was dropped.

client.query(birth={...}, sql="SELECT AVG(sign) AS v FROM planets")
# ValidationError: AVG requires a numeric column; sign is text

HAVING filters groups after grouping, and its operands may be aggregates. It requires a GROUP BY: some engines allow it without one and read the whole result as a single group, but in practice a bare HAVING is a WHERE written in the wrong place, and saying so is more use than answering it.

# Stellium: three or more bodies in one sign.
"SELECT sign, COUNT(*) AS n FROM planets GROUP BY sign HAVING COUNT(*) >= 3"

# Signs the chart moves slowly through.
"SELECT sign, AVG(speed) AS v FROM planets GROUP BY sign HAVING AVG(speed) < 0"

An aggregate collapses rows, so SELECT COUNT(*) FROM stars is one row and never meets the hundred-row cap. The cap counts the final result, after HAVING.

Astrological functions

Scalar functions, usable anywhere a column is: in SELECT, WHERE, GROUP BY, ORDER BY, and inside an aggregate. All are lookups or integer arithmetic; none of them calculates anything further.

RULER is the traditional rulership and only that: Mars for Scorpio, Saturn for Aquarius, Jupiter for Pisces. A modern variant would be a second answer to the same question decided by a setting the statement cannot show, so one query would mean two things; if it is wanted it will be a second function under its own name.

DIGNITY is the classical seven-planet table. Anything outside the seven is peregrine, including Uranus, Neptune, Pluto and the asteroids: inventing a domicile for them would be inventing astrology rather than reading it. Triplicity, term and face are real dignities and are deliberately not reported, because they need a degree and a day-night distinction and this function is given a sign. Mercury in Virgo holds both domicile and exaltation and is reported as domicile.

LENGTH and SUBSTR count characters, not bytes. A degree reads 11°1′36″, which is eight characters written in thirteen bytes, so LENGTH answers 8 and SUBSTR(dms, 3, 1) answers ° rather than the first half of it.

COALESCE earns its place on aggregates. No column of a chart is ever NULL, but an aggregate over a group that matched nothing is, so COALESCE(AVG(orb), 0) is the way to ask for a floor instead of an unknown.

NULL in, NULL out. A body with no house returns NULL from ANGULAR rather than a default. A name no sign or body carries is refused, and the message quotes the value it was given:

client.query(birth={...}, sql="SELECT ELEMENT('Lion') AS e FROM planets")
# ValidationError: ELEMENT does not know the sign 'Lion'

# Elemental balance.
"SELECT ELEMENT(sign) AS element, COUNT(*) AS n "
"FROM planets GROUP BY ELEMENT(sign) ORDER BY n DESC"

# Planets in their own sign.
"SELECT name, sign FROM planets WHERE DIGNITY(name, sign) = 'domicile'"

# The single tightest aspect in the chart.
"SELECT MIN(orb) AS tightest FROM aspects"

# Angular houses only, with the decan.
"SELECT name, sign, DECAN(longitude) AS decan "
"FROM planets WHERE ANGULAR(house) = 'angular'"

Subqueries

A subquery may stand after IN or NOT IN, after EXISTS, or as a single value in the SELECT list. The inner statement may name a different table than the outer one; both charts are already calculated, so this filters data in hand rather than calculating a second time.

# Everything aspecting a retrograde planet.
"SELECT a, b, aspect, orb FROM aspects "
"WHERE a IN (SELECT name FROM planets WHERE retrograde = 1)"

The inner statement must select exactly one column, and is told so by count:

client.query(birth={...}, sql="SELECT a FROM aspects "
                              "WHERE a IN (SELECT name, sign FROM planets)")
# ValidationError: subquery must select one column, got 2

The hundred-row cap applies to the final answer only. An inner query may legitimately produce all 953 stars while the outer answer is three rows.

NOT IN follows standard SQL where the inner set contains a NULL, which surprises people and is worth stating plainly: if any inner value is NULL, NOT IN returns no rows at all. The comparison is unknown rather than false, and unknown is not true, so nothing passes. IN is unaffected: it still matches the values that are there. Add WHERE col IS NOT NULL to the inner statement when the column has gaps. An empty inner set behaves the way the logic implies: IN matches nothing, NOT IN matches everything.

Two charts in one statement, which is synastry written as a join:

rows = client.query(
    birth={"date": "2000-01-01", "time": "12:00",
           "lat": 51.4779, "lon": 0.0, "timezone": 0},
    partner={"date": "1995-03-14", "time": "08:20",
             "lat": 41.0082, "lon": 28.9784, "timezone": 3},
    sql="SELECT a.name, a.sign, b.name AS partner "
        "FROM natal_a a JOIN natal_b b ON a.sign = b.sign LIMIT 50",
)

And transits against the natal chart:

rows = client.query(
    birth={"date": "2000-01-01", "time": "12:00",
           "lat": 51.4779, "lon": 0.0, "timezone": 0},
    moment={"date": "2026-08-17", "time": "12:00",
            "lat": 51.4779, "lon": 0.0, "timezone": 0},
    sql="SELECT t.name, t.sign, n.name AS natal "
        "FROM transits t JOIN chart n ON t.sign = n.sign LIMIT 50",
)

A join multiplies rows, so it reaches the hundred sooner than a plain select. Both of these overrun it without the LIMIT.

Rendered output

Without format the answer is rows. box and csv render it as text, meant for a terminal rather than for parsing, and arrive as str:

print(client.query(birth={...}, sql="SELECT name, sign FROM planets LIMIT 3",
                   format="box"))
# ┌─────────┬───────────┐
# │ name    │ sign      │
# ├─────────┼───────────┤
# │ Sun     │ Capricorn │
# └─────────┴───────────┘

border picks the box style from sharp, round, ascii, rules and markdown.

json renders too, and the client parses it back, so it reads like any other response rather than like a string you have to decode yourself.

scalar returns the single value a one-row, one-column answer holds, and object returns a one-row answer as a mapping, which saves indexing into a list of one:

client.query(birth={...}, sql="SELECT COUNT(*) AS n FROM stars",
             format="scalar").data          # 953.0

client.query(birth={...}, sql="SELECT name, sign FROM planets LIMIT 1",
             format="object").data["sign"]  # 'Capricorn'

Every endpoint

Every published operation, 66 of them, has a method on both clients and a module-level shortcut, all generated from the schema: natal, transits, synastry, composite, solar_return, progression, harmonics, astrocartography, vedic_chart, zodiacal_releasing, firdaria, horary, electional and the rest.

Anything new is reachable before this client names it:

result = client.request("POST", "/v1/harmonics", json={"birth": {...}})

Authentication, timeouts, retries and error handling behave identically there.

Errors

from astromansion import QuotaExceededError, RateLimitError

try:
    chart = client.natal(date="1990-07-19", lat=41.0, lon=29.0)
except RateLimitError as error:
    print("wait", error.retry_after, "seconds")
except QuotaExceededError:
    print("this period's allowance is spent")

All descend from AstroMansionError. Each carries status_code, error_code, details, request_id and retry_after when the API supplies them.

Rate limits and quota

A rate limit clears on its own after retry_after. A spent quota does not: it needs a new period or a larger plan. They are separate exceptions for that reason.

The client retries only failures that carry no result: connection errors, 429 and 5xx, twice by default, honouring Retry-After. A refusal you must fix is never retried.

client = AstroMansion(timeout=60.0, max_retries=0)

Documents

pdf = client.export_pdf(date="1990-07-19", lat=41.0082, lon=28.9784)

with open("chart.pdf", "wb") as file:
    file.write(pdf)

Or name a path and let the client write it:

client.export_pdf(date="1990-07-19", lat=41.0082, lon=28.9784, output="chart.pdf")

Every endpoint that answers with a document takes output the same way: export_pdf, export_csv, render_svg, render_png, render_biwheel and render_sharecard, on both clients.

client.render_svg(date="1990-07-19", lat=41.0082, lon=28.9784, output="wheel.svg")

Nothing is written to disk unless you name a path, so a call cannot overwrite a file you did not choose. Without one you get the bytes and decide yourself.

Security

The key travels in the X-API-Key header, never in a URL. It is masked in repr(client) and appears in no exception or log line the package writes. Keep it in the environment or a secret store, not in source control. Rotate it from your account page if it leaks.

Staging

client = AstroMansion(base_url="http://localhost:8000")

Also readable from ASTROMANSION_BASE_URL.

Links

• API reference
• Documentation