The AI in Astrolium has a name, and the name is Sky. It builds natal, transit, synastry, return, and forecast readings from the chart's own computed facts, in six languages, with citations back to the positions it read from. The output follows your method, because Sky reads every chart against the rules, instructions, and notes you load into your Knowledge Base.
The free chart generator casts a chart in 15 seconds. The AI in astrology guide covers how to ground a model in your own material. For the predictive layer Sky reads from, see transits and predictive timing. For the written deliverable, see Report Studio. Sky starts on the $11 per month Pro plan; see pricing.
What Sky actually does
There are two failure modes for AI in astrology, and Astrolium tries to avoid both. The first is the ChatGPT failure: a confident, generic, slightly wrong interpretation that does not know which house system you use, which planets your school treats as rulers, or even which chart it is reading. The second is the no-AI failure: 90 minutes of writing per client per session, much of it spent on scaffolding the practitioner has written 200 times already.
Sky sits between those two. Your material is the input, so the interpretation arrives in your method and your register. Your time goes to the part that needs a human: the bridge between the chart and the client's life.
The Knowledge Base grounds every reading
Upload your interpretation rules, your house-by-house instructions, your source authors, and the notes you have built over years of reading. Sky reads each chart against that material before it writes a word. A practitioner who weights sect heavily gets sect-led readings. A practitioner who never uses outer-planet rulers never sees them.
Every output opens in a markdown editor with citations attached, so when a client asks where a claim came from you can point at the wheel. Citations are sticky, so editing a sentence keeps the link. You can display them inline, in brackets, after the paragraph, or not at all, and you can ask Astrolium to flag any answer that is not grounded in a source.
When Sky reads something the way you would not, Refine saves your interpretation as a note in your Knowledge Base, scoped to everything, to one client, or to that client's charts. Sky leads with your wording wherever it applies. That is the mechanism by which the assistant gets more like you over months rather than staying frozen at whatever it knew on day one.
Memory, and how to see all of it
Sky keeps two kinds of memory, and the separation is the point.
Client Memory holds short, stable facts about one client: what they do, who is in their life, what they came to you for. It is shared only with people who can already read that client's notes, so it never leaks across a team.
Practice Memory holds how you work. Doctrine, technique, workflow, interpretive style, writing style, and exclusions get their own categories. It is private to your account.
Each scope runs in one of three modes: save automatically, ask before saving, or do not use memory. Turning a scope off stops both reading and saving, and the entries you already have stay where they are until you delete them. There is a manager for each: read every stored line, edit it, disable one without deleting it, add one by hand, delete one, or clear the lot. Nothing is inferred behind your back and nothing is mined out of old transcripts.
Conversations are a separate thing. They are saved, grouped, and resumable, kept for 6 months after the last message, and deletable one at a time. Deleting a conversation leaves your notes and your memory intact.
Every module has a Sky answer
Sky started as a chart-panel assistant. It is now a workspace assistant, and the difference shows up when you ask a returns question from the Horary tab and get an answer instead of a shrug.
The core tools cover natal snapshots, relationship charts, return charts, sky events, and transit-window searches. The specialist tools go further: time-lord techniques including profections, Zodiacal Releasing, Firdaria, and Decennials; the Arabic lots with their formulae; midpoint dial geometry; the astrocartography line inventory; Human Design charts and two-person interactions; and numerology. A coverage registry backs this with a build-time test that fails if an active module ships without a declared Sky path.
Two of these deserve naming because nothing else does them. Ask Sky where to spend a birthday year and it ranks real places against one or two named life goals through astrocartography, then calculates a relocated solar return for the top candidates; if either calculation fails, it returns nothing rather than a plausible guess. Ask it to pick a date for signing a contract and it runs an electional search across the window you name, up to 730 days, and returns ranked windows with their factors and cautions.
Horary is the one place Sky deliberately stops short. It recognises a horary question in chat and offers you the route into the horary workspace, but it cannot press Cast, and it will not report a chart as cast on the strength of a stray "yes" in a conversation.
Ask AI, one row at a time
The chat panel in the right rail is not the only way in. Ask AI buttons sit beside individual pieces of data across the workspace: a single transit row, a midpoint structure, a house cusp, an aspect, a planet on the wheel, an astrocartography line, a place, an electional window, the time navigator itself.
Each button sends that row's actual computed evidence and carries an accessible label naming its subject, so a screen reader announces which transit you are asking about rather than "button." In practice this is how most questions get asked: you are mid-reading, one line looks odd, and you want a second opinion on that line and nothing else.
When a calculation fails, it says so
The most important trust work of the last quarter was not a new capability. It was making Sky honest about its own machinery.
A running answer shows a checklist of the tools being called: casting the natal chart, finding transit dates, calculating time-lord periods. Each entry resolves to done, needs more input, or unavailable, with a specific next step where one exists. Ask for a return without a location and it asks for the location rather than inventing one. Ask for something outside a tool's contract and it refuses instead of pretending to process it. If an answer hits the length limit, it is labelled as cut off, offers Continue, and still saves to history rather than vanishing.
None of that is glamorous. It is the difference between an assistant you can put in front of a paying client and one you have to check afterwards.
Guard-rails
Some limits are enforced server-side on every request, and workspace settings can add care but never remove it.
Sky does not predict death or any certain fatal outcome. On suicide or self-harm it responds with support, urges urgent local or emergency help, and stops the predictive reading. It reads only the client data you actually supplied, and never invents a name, a contact detail, or a fact you did not give it. Birth-time caveats are binding: when a calculation comes back with an unreliable time or a Moon in a range, the answer says so and does not rebuild the missing houses, angles, sect, or rulers from context. Medical, legal, financial, pregnancy, and fertility themes stay available, framed as astrological judgment rather than clinical, legal, or financial instruction.
Three further rows are listed in the guard-rails screen as not yet enforced rather than promised: third-party charts without consent, named public figures, and breakup phrasing.
The settings screen lists every row and labels it always on or not yet enforced, so you can see which is which before you rely on it.
The interpretive lens
Astrology is not one tradition. Sky reasons in one of four lenses: follow the open chart, modern psychological (Jung, archetypes, growth-oriented), traditional and Hellenistic (sect, dignities, lots, time-lords), or Vedic (Jyotish logic, dashas, karakas). The lens shifts rulerships, reading order, and vocabulary. Mars rules Scorpio in the traditional lens; Pluto does in the modern one. Hellenistic starts with sect and the luminaries; modern psychological starts with the Sun, Moon, and Ascendant.
Set it per chart or as a workspace default, and run a traditional natal in the morning and a modern synastry in the afternoon without reconfiguring anything. The same setting decides the register of the prose in Report Studio documents.
Languages
Three language lists run through Astrolium, and they are not the same list. The interface ships in English and Russian. Sky replies in six: English, Russian, Spanish, German, French, and Italian. Report Studio writes documents in its own seven, which adds Ukrainian and Brazilian Portuguese and drops Italian.
Sky writes natively in each language rather than translating, with the astrological vocabulary that language actually uses, and you choose whether it keeps traditional terms or explains them in plain words. Reply language can follow the interface or be set on its own, which is what you want when your workspace is in one language and your client reads another.
Bring your own key
On Adept and Master you can run Sky on your own provider account. Three providers are supported: OpenAI, Anthropic, and OpenRouter. Pick the model yourself, and the messages stop counting against Astrolium credits entirely, because you are paying your provider directly.
Keys are encrypted at rest in Supabase Vault and never displayed again after you save them. Free and Pro use the built-in managed model.
Versus other AI in astrology
| Tool | Knows your chart? | School-aware? | Citations? | Memory you control? |
|---|---|---|---|---|
| ChatGPT (raw) | No | No | No | Opaque |
| Co-Star app | Partial | No (modern only) | No | None |
| Generic astrology AI tools | Partial | No | No | None |
| Sky in Astrolium | Yes (Swiss Ephemeris) | Yes (4 lenses) | Yes (linked to wheel) | Readable, editable, deletable |
The competitive question is not whether a tool uses AI. It is whether the AI reads your way. A prompt that says "interpret this natal chart" gets a generic answer because the model has no chart and no method in front of it. Sky's prompt carries the actual positions, the house system, the rulers in use, the interpretive lens, your Knowledge Base, and whatever you have told it about this client.
Pricing
Sky starts on the $11 per month Pro plan with 50 AI credits a month. Adept at $29 raises that to 300, and Master at $79 to 1,000. Credits can be topped up with a one-time $9 purchase on any plan, including Free. On Adept and Master, bringing your own key makes AI messages unmetered. The free tier has no Sky chat; cast unlimited natal charts for free and add the assistant on Pro. See pricing for the full breakdown.
Continue exploring
- Chart generator — cast a natal in 15 seconds, free
- AI in astrology guide — how to ground a model in your own method
- Report Studio — the written deliverable Sky drafts into
- Transits — the time navigator Sky reads windows from
- Solar Fire vs Astrolium — the desktop comparison (no AI ships in Solar Fire)
- AI interpretation vs manual — where the AI helps and where it stops
- Rectification methods compared — how the Rectification Wizard scores candidate times