AI Life Chat

The panels above say what AI Life Chat does and what it needs. This long read goes underneath them: what the classical texts say about putting a question to a chart, what the code does between your question and the reply, how to read every kind of reply, one sample chart asked a series of real questions, and what a wrong or missing birth time does to the answers. If dashas are new to you, what a Mahadasha actually is is the six-minute introduction.

Every reply and number quoted below came from running the code that serves the chat, on the sample person the free calculator pages share - born 14 August 1994 at 06:42 IST in Ahmedabad (23.0225° N, 72.5714° E) - asked at noon IST on 5 October 2026. It is not a real person. Where the engine declines to answer, this page shows that too.

Putting a question to a chart: what the texts describe

Indian astrology has a branch built for questions. Prashna, or horary astrology, casts a chart for the moment a question is asked and reads the answer from that chart alone. Daivajna Vallabha, the text behind this app's Prashna feature, opens with the conduct of the exchange: the query should be put with a settled mind, and put once (Ch. 1, Sl. 9-10), and the astrologer should weigh its very syllables, along with the surroundings, before speaking (Sl. 11). AI Life Chat is not prashna. It never casts a chart for the moment you type. It reads the birth chart saved on your profile and the dasha running today - the natal method of Brihat Parashara Hora Shastra (BPHS). The Prashna calculator runs the question-moment method without an account.

What the two methods share is the first move: decide which house the question belongs to. Daivajna Vallabha Ch. 2 lists the matters judged from each house and tells the reader to treat the house of the matter in hand as the starting point (Sl. 18). BPHS Ch. 11 Sl. 2-13 gives Parashara's list for a birth chart. The chat uses one fixed table, TOPIC_HOUSE in aiEngine.js, shared with the Decision Engine so that a topic means one house everywhere in the app.

Two rows deserve a second look. Both texts put trade itself in the 7th house; the 11th is income and the gains from an enterprise, so a business question here is read from what a venture brings in, not from the dealing. And both treat the 6th as the house of illness and opposition, while the body and its vigour belong to the 1st. The engine labels its 6th-house reading "health and daily routine": a health question is read from the house of strain, not of constitution, and nothing the chat says on health is medical advice. Each house has a learn page: the 10th, 2nd, 7th, 6th, 11th and 1st.

Once the house is chosen, the texts judge it several ways at once. BPHS Ch. 11 Sl. 14-16 says a house does well when a benefic joins or aspects it, or when its lord is in a good state, and suffers when its lord is with a malefic or beaten in a planetary war. Daivajna Vallabha Ch. 2 Sl. 2 and Saravali Ch. 3 Sl. 25 both look for the lord's own aspect on its house. BPHS Ch. 27 Sl. 26-29 builds a full house strength, Bhava Bala, and adds the lord's strength to it. The chat's verdict takes that last ingredient alone - the lord's strength, as this app's simplified twenty-point Shad Bala - on one scale for all seven grahas, where BPHS sets each its own bar (Ch. 27 Sl. 32-33). It is a narrower test than any of the three texts describes.

The other half of the verdict is the period. BPHS Ch. 47 Sl. 2 separates a dasha's general effects, which come from the planet's nature, from the distinctive ones, which come from its placement. Sl. 5-6 gives the test: favourable when the dasha lord is in the Lagna, exalted, in its own sign or a friendly one; unfavourable when it is in the 6th, 8th or 12th, debilitated or in an enemy's sign. Saravali Ch. 41 Sl. 1 and 6-7 judge a period by much the same natal conditions, and Daivajna Vallabha Ch. 1 Sl. 3-8 counts a planet in a friend's sign among the states that deliver what was asked about. The engine is stricter: in its three-step quality a friendly sign is only neutral.

One thing the texts ask for that the chat does not do. Advanced Predictive Astrology (Vol-II, pp. 229-230) has the reader judge the dasha and sub-period lords first and the transits second, comparing the dasha to the hour hand of a clock and the transit to the minute hand. The chat's verdict reads the hour hand only. For the minute hand, the Gochar calculator shows today's transits from your Moon and Muhurat scores specific windows.

What happens between your question and the reply

The whole exchange is one request to the server, handled by send() in chatController.js. Nothing in it is random and, as shipped, nothing leaves the server.

1. 1.The question is checked: an empty message is refused, and so is anything over 500 characters.
2. 2.Your memory is read: your last 60 questions, taken before the new one is stored, for recall, and your 50 most recent Life Memory entries, for patterns. If either read fails, the reply is still written without that sentence.
3. 3.The chart and the numbers are recomputed from your saved birth details for the moment of asking, and the running Mahadasha and Antardasha are looked up on the Vimshottari timeline.
4. 4.intent.js reads the question: its kind, the life area it names, and the state it was typed in.
5. 5.The Decision Engine, decide() in decisionEngine.js, runs for the question's category - for every question, though only a decision reply quotes it.
6. 6.The composer, composeReply() in chat/composer.js, chooses which facts to say and in what order. Its header comment is blunt: it computes nothing.
7. 7.Question and reply are stored with the intent, voice, verdict and follow-up recorded for them.

How a question is read

The reading is by keyword, on one message, after lower-casing it and stripping everything that is not a Latin letter, a digit or a space. Hinglish is on the lists throughout - "kab", "karu", "naukri", "shaadi", "paisa", "pareshan". The intent is decided in a fixed order, and the first rule that matches wins.

The order is deliberate. "Should I quit?" and "When should I quit?" both carry a decision marker, but the second opens with "when", so it gets a date, not a verdict. A side effect: unless an earlier rule claims it, any message containing "should I", "decide" or "decision" is read as a decision - even an open one such as "What should I focus on this month?", which then receives a verdict for the general house. The life area is read separately from five word lists, checked in the order career, money, love, health, business; if none matches, the area stays empty and is not guessed. A question with neither a recognised intent nor an area is marked low confidence, and the composer then asks instead of asserting.

The same step reads the state the message was typed in, from distress and confusion words. The result picks one of three voices - steady, guiding or direct - and nothing else. It never becomes a statement about you: the code's own note is that being told what you feel by software is irritating when it is right and insulting when it is wrong.

The verdict and its arithmetic

The Decision Engine page sets out the rule: up to 60 points for the quality of the running Mahadasha, up to 40 for the Shad Bala of the ruler of the question's house; YES at 65 and above, WAIT from 35 to 64, NO below. Quality has three steps. It is challenging when the Mahadasha lord sits in the 6th, 8th or 12th house or in a debilitated or enemy sign; good when, outside those cases, it is exalted, in its moolatrikona or in its own sign; neutral otherwise. Rahu and Ketu have no dignity entries in the engine's table, so their periods are neutral or challenging, never good.

Two consequences. The verdict mostly follows the Mahadasha; the house ruler moves the confidence inside a band and only rarely moves the word. And neither input changes from week to week: ask the same decision question next month, inside the same Mahadasha, and the verdict and the number will be the same. A verdict here describes a period of six to twenty years and one house of your chart, not this week.

Two word lists behind one reply

One detail of the plumbing explains some replies. The phrase "setting the terms for career here" uses the area read by intent.js. The verdict uses a second, older classifier in aiEngine.js, which counts keyword hits in its own five lists and takes the list with the most. The two overlap heavily but not completely, and neither understands a word form it was not given.

The practical rule: name the area with a plain noun - job, money, marriage, health, business - and read the reasoning note, which states how the question was taken. If it does not match what you meant, rephrase; the engine answered a different question.

Dates, numbers, patterns and recall

Every date in a reply is the end of your running Antardasha, taken from the timeline. BPHS Ch. 51 Sl. 1-2 sizes a sub-period as the product of the two lords' years divided by 120, the first one belonging to the Mahadasha's own lord; the full table for your chart is on the Vimshottari Dasha calculator. The composer treats a boundary within 60 days as "soon": only inside that window does a timing reply add a day count, or an emotional reply or a YES gain a dated sentence of its own. Dates are printed as day and month in Indian Standard Time, without the year. A sub-period can run for over three years, so check the year on the Timing page before planning around it.

Decision and timing replies carry one sentence naming your Mulank and Personal Year. Mulank is the day of birth reduced to one digit - the "Moolank" or psychic number of Advanced Vedic Numerology. Personal Year is birth day plus birth month plus the current year, reduced to one digit, as in the Numerology Professional Course Notes (p. 108) and Numerology Metaphysical Secret. The sentence is fixed wording. The engine does not test whether the numbers agree with the dasha verdict, so read it as two values placed beside the answer, not a second opinion. The two systems are compared properly on the Convergence tool, and When astrology and numerology disagree explains why a disagreement is worth noticing.

A pattern, to this engine, is a count. Each Life Memory entry is stamped with the Mahadasha and Antardasha lords running on the date it happened. Entries are grouped by tag, by that pair of lords and by type, and a group becomes a pattern at three entries. A pattern question is answered only from those groups. A decision or timing reply quotes a pattern only when it is in the same life area as the question, and never after a refusal.

Recall is narrower still. Each question is stripped to content words - a stop list removes "should", "I", "my", "the" and their Hinglish equivalents - and compared with your own questions from the last 90 days. A match needs at least two shared content words, making up at least half of the shorter question. The engine's earlier replies are never searched: its own words are not a memory of you.

Last, the wording. Each voice has three openers and three closing lines, and the choice is made by a hash of your question's text, not by chance. The same question always produces the same wording, because a reply rephrased on a second asking would look like a change of mind.

One chart, asked several questions

The sample has Leo rising at 2.33° and the Moon at 24.64° Libra in Vishakha, a nakshatra of Jupiter. The Moon had crossed 34.8% of it at birth, so 65.2% of Jupiter's 16 years remained - 10.43 years, to January 2005. Saturn's 19 years followed, and Mercury's 17 run from 19 January 2024 to 18 January 2041. Inside them the Ketu Antardasha lasts 17 × 7 ÷ 120 = 0.99 years: 16 June 2026 to 13 June 2027. The Mulank is 5 (14 reduces to 5) and the Personal Year for 2026 is 5 (14 + 8 + 2026 = 2048; 2 + 0 + 4 + 8 = 14; 1 + 4 = 5). The side panel reads Mahadasha "Mercury · till 2041", Antardasha "Ketu", Moon "Libra · Vishakha", Ascendant "Leo", and an energy of 3.1/10 for the day.

Mercury sits at 28.35° Cancer in the 12th house, in an enemy's sign and combust. The house or the sign alone would be enough: the Mahadasha quality is challenging, worth 0 of 60. BPHS reads the placement the same way - Ch. 47 Sl. 62-65 gives Mercury's period its good results from exaltation, own or friendly signs and the 5th, 9th and 11th houses, and Sl. 66-70 withdraws them in the 6th, 8th or 12th. See Mercury in the 12th house and the Mercury Mahadasha. With nothing from the period, every verdict on this chart until 2041 rests on the house ruler alone.

Asked "Should I switch jobs this quarter?", the engine reads a decision about career and shows the NO badge. The reply states the period, quotes the Decision Engine's reasoning word for word - the Mahadasha reads challenging; Venus, ruler of the career house, is weak at 26% of maximum Shad Bala - adds the numbers sentence, and gives the verdict: "So no. Not this one, and not on these terms." Because it is holding you back, it names the date: hold the irreversible part until 13 Jun and use the gap for the reversible parts. The reasoning note reads: "Read as a decision question about career. Composed from the fact that you are in a Mercury Mahadasha with Ketu running underneath it, your Personal Year 5, and this app's own decision verdict (NO, confidence 11/100)."

One honest reading of that date. 13 June 2027 is the next boundary on this timeline, not a date on which the verdict changes: asked again on 14 June 2027, the question would get the same NO at 11. BPHS Ch. 52-60 give every sub-period its own reading, so the texts do expect the tone to shift there; the chat names the date and quotes none of those chapters.

"Should I propose to my partner?" comes back WAIT at 35 - exactly on the line. Saturn rules the 7th and holds 17.63 of 20 points; 88.15% of 40 is 35.26, which rounds to 35, the lowest score that reads WAIT. The verdict sentence changes with the word: "So not yet. This is a wait, not a no." A verdict on a threshold is one to treat as undecided; the birth-time section shows this one flip on eleven minutes.

"When should I quit my job?" opens with "when", so it is a timing question and carries no badge. The reply says the sub-period runs to 13 Jun, that the date is on the timeline and not a guess, and - the boundary being more than 60 days away - that the pressure does not change much inside it, so the moment should be picked on practical facts. Its reasoning note does not mention the decision verdict, although one was computed: the note describes only the reasoning the reply used.

"Why do I keep repeating the same career pattern?", with nothing logged, gets the plainest reply in the engine: nothing has repeated often enough to call a pattern, and it takes three of a kind under the same period before a repeat is more than a coincidence. With three career decisions logged under Mercury / Ketu, the same question quotes that group, and the job-switch reply gains a sentence: "Your own log backs this up: 3 career decisions under your Mercury / Ketu period." "I feel stuck and anxious about everything" switches the voice to steady ("Take a breath first."), carries no verdict, and offers one action: the smallest thing that can be finished today. "What does my chart say?" matches no list, so the reply states the period and asks a question back.

Recall was tested with "Should I switch jobs this quarter?" stored sixteen days earlier and "Should I switch jobs now or wait?" asked today. Two of three content words are shared - 0.67, over the bar - and the reply adds "You asked something close to this 2 weeks ago". Against "Will my job improve?" the stored question scores zero: "job" and "jobs" are different strings to this matcher.

Now the same chart on 1 December 2023, seven weeks before Saturn's Mahadasha ended. Saturn sits in Aquarius, its moolatrikona sign, in the 7th house: quality good, 60 points. The career question scores 60 + 11 = 71 and the reply reads "So yes - this is a reasonable time to move on it." Every area was a YES that day - 71, 72, 95, 95, 72 and 85. With the Jupiter sub-period ending within 60 days, the reply added the dated sentence: 19 Jan, about 49 days out. It did not say that the Mahadasha ended the same day, and that every score on this chart then fell by 60 points.

That table is the most useful thing on this page for reading a verdict. The instrument is coarse: it can tell a nineteen-year stretch from a seventeen-year one, and it cannot tell October from November.

How to read each kind of reply

• A YES, WAIT or NO badge is the Decision Engine's verdict for the house the question was matched to. The confidence in the reasoning note is the 0-100 score itself - 60 for the period, 40 for the house ruler - not a probability that things go well.
• A WAIT is not a soft NO. It is the 35-64 band: in a challenging Mahadasha it means an unusually strong house ruler, in a neutral one it is the default.
• A refusal: with no birth time the Decision Engine returns a confidence of 0, and the reply says it cannot give a yes or a no, names what is missing and stops.
• A timing reply has no badge and names no good month, because the engine has no rule that produces one. A pattern reply quotes a counted group from your own log or says there is none.
• An emotional reply names the period and offers one small action. It is not a diagnosis, and not a substitute for talking to someone who can help.
• A general reply states where you stand: period, Moon sign and nakshatra, Mulank and Personal Year. The five topic chips send a question of this kind. If the reply opens by saying it wants to answer what you actually asked, the question matched no list.
• The reasoning note is the audit trail: how the question was read, which facts were used, and who wrote the wording. If it says a language model wrote it, your deployment has that path switched on.
• The follow-up is a button. Tapping it sends that sentence as your next message, which the built-in engine reads like any other; answering in your own words, with the area named, gets a more specific reply.
• "Save this" stores the reply and its note among your saved insights. "Track this decision" appears only on replies with a verdict and writes a decision entry to Life Memory, tagged with the area - which is where the pattern counts get their rows.
• In the side panel, a dash or "Birth time needed" is a missing measurement, in a muted colour so it cannot pass for a value. Emotional state and active themes appear only once you have logged something; they are counts from your entries, not readings of you.

Birth time: what moves and what does not

Two things in a reply depend on the birth time, differently. The dates slide. The Vimshottari clock is set by how far the Moon had travelled through its nakshatra at birth, and the Moon moves about half a degree an hour. On the sample chart, whose opening lord is Jupiter at 16 years, each minute of birth time moves every later boundary by about four days: born at 06:52 instead of 06:42, the Ketu sub-period ends on 1 May 2027 instead of 13 June. Born at 07:42, the person is in Venus's sub-period on 5 October 2026, running to July 2029 - a different sub-lord and a different date in the reply.

The verdicts jump. They depend on the houses, the houses on the Ascendant, and the Ascendant changes sign about every two hours. The sample's is only 2.33° into Leo. Eleven minutes earlier, at 06:31, it is 29.87° Cancer and every house moves by one sign. Mercury is then in the 1st house, still in an enemy's sign, so the period stays challenging; the rulers change underneath it. Saturn still rules the 7th, now from the 8th house, and its strength slips just far enough to take the love verdict from WAIT at 35 to NO at 34. Health, now ruled by Jupiter, falls from 35 to 7. The Lagna calculator shows how close to a sign edge your own Ascendant sits.

With no birth time the app casts the chart for 12:00 local time. For the sample that is five hours and eighteen minutes late; the Moon is 3.1° further on, and Mercury's Mahadasha is dated April 2020 to April 2037 instead of January 2024 to January 2041. There are no houses, so a decision question is refused. A timing question is still answered with a date - "That sub-period runs to 24 May" - and the reply does not flag that it rests on an assumed noon. Treat any date from a profile without a birth time as approximate to months or years, not days. What a missing birth time costs lists what survives, and why apps disagree on your nakshatra covers charts near a boundary.

There is a floor under this that no birth certificate fixes. The Moon comes from a Meeus series accurate to about 0.1° to 0.3° by the project's own notes; it is not Swiss Ephemeris. For a 16-year opening lord, 0.1° is about six weeks on every later date and 0.3° about four months, so a boundary within a few weeks of today deserves the least weight. The methodology page sets out how the engines were checked.

What the chat does not do

• As shipped, no language model writes the reply. The code carries a second path: a deployment that sets CHAT_ENGINE to "llm" and supplies a Groq or Anthropic key sends the question, up to six earlier turns, the computed chart context and your five most recent Life Memory entries to that provider, under a fixed prompt that forbids inventing placements and requires the Decision Engine's verdict unchanged. The reasoning note says which one wrote the wording. The privacy policy lists exactly what would be sent, and the data protection notice names the configured vendor by asking the running server, or says there is none.
• It matches words; it does not understand language. Questions typed in Devanagari match none of the lists and are read as general. Roman-script Hinglish is recognised. The reply is composed in English whichever language the app is set to.
• The built-in engine does not carry a conversation. Each message is read on its own words; a bare "why?" is a new general question.
• Its verdict does not read transits, sub-periods, yogas, doshas, aspects on the house or divisional charts. Most of those are computed on the Kundali page; the built-in reply does not quote them.
• It does not predict events. The only dates it can print are boundaries on your dasha timeline.
• It prescribes no remedies, gemstones, mantras or rituals, and gives no medical, legal or financial advice. A YES on a money question reads one planet's period and one house ruler; it is not a recommendation to invest.
• If paid plans are switched on, the free tier allows two replies a day, counted on the UTC date, which turns over at 05:30 in India; a question past the limit gets a fixed notice, not a reading. As the product ships today, every account is unlimited.

What is stored, and how to remove it

Each question and reply is stored on your account: the text, the reasoning note, the topic and, for replies, the intent, voice, verdict and follow-up. Memory is kept per profile, so a question about another saved profile is never answered against your own history. "Clear chat history" asks for confirmation and then deletes every chat message on the account - every profile's conversation, not only the one on screen - and leaves your chart, birth details and account as they are. It does not remove replies you saved or decisions you tracked; those live in your saved insights and in Life Memory. Deleting the account itself is done in Settings. The privacy policy and the data protection notice say how birth details are handled.

Where the chat connects to the rest of the app

The chat is a front door onto engines that have their own pages. The verdict is the Decision Engine's, and that page adds what the chat leaves out: the two axes drawn separately, a risk map, and a what-if box that scores the same decision at the dasha handovers ahead. The period it names is laid out in full on Timing, whose action windows use the same scorer for the general question. The chart underneath both is the Kundali page. For a specific day and hour, Muhurat is the tool the timing reply sends you to; for a question answered from the moment of asking, Prashna. The energy figure in the side panel belongs to the daily dashboard, the two numbers in the reply to Numerology, and the record that lets a verdict be checked against what happened to Accuracy.

Without an account, the pieces can be checked one at a time: the birth chart tool for the chart, the Shad Bala calculator for the strength of each house ruler, the Moon sign and nakshatra calculators for the two values a general reply quotes, and the numerology profile for the Mulank. The Personal Year calculator shows the calendar-year figure the chat quotes beside a second, Vedic personal year that runs birthday to birthday, and Personal Year, explained covers the arithmetic. Antardasha and Pratyantar, explained covers the sub-periods, and the books behind every engine are listed under sources.

Questions people ask about AI Life Chat

Is a language model writing my answers?

Not as the product ships. Every figure is computed by the app's own engines and the sentences are assembled by a rule-based composer from fixed phrasings; the reasoning note under each reply says so. If a deployment routes the same facts to an external model, the note says a language model wrote the wording, and the data protection notice names the vendor.

I asked the same decision question a week later and got the same verdict and number. Is it stuck?

No. The verdict is the quality of your running Mahadasha plus the Shad Bala of one house ruler. Neither changes inside a Mahadasha, so the answer holds until that period ends. For a specific day, use Muhurat.

My Timing page shows one colour and the chat said something else about my job. Which is right?

Both, about different houses. Timing scores the general question, which is the 1st house. The chat scores the area it reads from your wording, so a job question is scored on the 10th house and its ruler. The Mahadasha term is identical in both.

Can I ask in Hindi?

In Roman script, yes: the word lists include Hinglish such as kab, karu, naukri, shaadi, paisa and pareshan. A question typed in Devanagari matches none of the lists and is read as a general question. The reply is composed in English either way.

I do not know my birth time. What can the chat still tell me?

Your Mulank and Personal Year are unaffected, and your Moon sign usually is too. A decision question is refused, because the houses cannot be computed. Timing replies still come with dates, but those are cast from an assumed noon and can be out by months or years. This article lists what survives and how people recover a time.

Why did my marriage question get a reply about "this" and not about relationships?

The area is read by keyword. "Marriage", "partner", "shaadi" and "rishta" are on the list; "married" is not. When no area matches, the reply speaks generally and the verdict is scored on the 1st house. Check the reasoning note and rephrase with the noun.

Can it tell me when I will get a job or get married?

No. The engine has no rule that produces the date of an event. The only dates it prints are the boundaries of your dasha periods, described as changes of period, not as promises.

How is this different from the Prashna feature?

Prashna casts a chart for the moment you ask and reads the answer from that chart, following Daivajna Vallabha. The chat reads your birth chart and your running dasha. The same question can get different answers from the two; they are different methods looking at different charts.