One conversational endpoint
Your users ask in their own words. Vedika computes the chart in code first, derives the placements, dashas and yogas deterministically, builds the prompt out of those computed facts, and only then calls a model. The chart it used comes back in the same response, so you can check the answer against it.
The whole surface
POST https://api.vedika.io/api/v1/astrology/query
Authentication
X-API-Key: your key
The smallest call that works
curl https://api.vedika.io/api/v1/astrology/query \
-H "X-API-Key: $VEDIKA_KEY" \
-H "Content-Type: application/json" \
-d '{
"question": "What does my chart say about career?",
"birthDetails": {
"datetime": "1990-05-15T10:30:00+05:30",
"latitude": 28.6139,
"longitude": 77.2090,
"timezone": "Asia/Kolkata"
}
}'
question is the only required field.
Everything else, birth details included, is optional — a question
with no chart is answered as theory.
One endpoint
A question in plain language. There is no route to pick, no chart
type to choose and no second call to stitch on —
question is the one required field,
and birthDetails,
system,
language and
systemPrompt are all optional.
POST /api/v1/astrology/query
Grounded
The chart is computed at step 2 of a twenty-three-step pipeline. The model is not called until step 9. Everything in between — placements, dashas, yogas, doshas, the classical lookups — is derived in code, and the prompt is built from the result.
Chart at step 2, model at step 9
Four systems
Ask across Vedic, Western, KP or Vastu on the same endpoint, and get the answer back in any of 30 accepted language codes — 15 of them Indic, in native script.
vedic · western · kp · vastu
Plug it in
Steps 1 to 3 run against the keyless sandbox and take a couple of minutes. Step 4 is the only one that needs a key, and the request body does not change when you swap it in — that is the whole point of having a sandbox that mirrors the production shape.
01
/sandbox/* is mounted outside the
auth stack: no key, no wallet, no billing. It answers with static
sample data in the production response shape, and the sample text
says so in its own body.
curl https://api.vedika.io/sandbox/chat/ask \
-H "Content-Type: application/json" \
-d '{"question":"What does my chart say about career?"}'
02
It is one POST with a JSON body, so fetch
is a legitimate answer. The SDKs exist to give you types, retries
and a base URL you do not have to remember. See
the SDK section for
the four published packages and their real signatures.
03
response is the prose,
birthChart is the chart it was
written from, and followUpSuggestions
is your next row of chips.
Rendering the response
covers all seven fields and the whitespace rule that catches people out.
04
Change /sandbox/chat/ask to
/api/v1/astrology/query, add
X-API-Key, and send the birth
details you were going to send anyway. That is the entire
difference between the two calls.
An honest note on "five minutes". Steps 1 to 3 really are a few minutes' work. Step 4 is not, because it needs a paid account — there is no free production tier, no trial and no credit grant, so the clock on a first production call includes signing up and loading a wallet. The integration is five minutes; the procurement is not, and we would rather say so.
The query path
"Grounded" is an ordering claim, not an adjective, so here is the order. This is the real sequence a request runs through, with the real step numbers the pipeline itself uses. The model appears at step 9 — after the chart, after the deterministic interpretation, after the prompt has been assembled out of computed facts. Steps are skipped when they do not apply; none are ever reordered.
0a · 0.2
A zero-tolerance hard block runs first, then crisis tagging with multi-turn escalation. These are defence-in-depth and they run before a single planetary position is calculated, so a request that must be refused never touches the ephemeris or a model.
01
Meta questions ("what can you do?"), greetings and safety redirects are answered here and return immediately. They are not worth a chart and they are not worth a model call.
02
Positions and transits are calculated from the birth details in the request. This is deterministic arithmetic against an ephemeris — the same inputs give the same chart every time, and it happens before any prompt exists. If a long conversation needed summarising, that summary was produced just above this line and it is also deterministic: no model has been called yet.
03
Deliberately before retrieval, so the detected topic can filter what gets retrieved rather than the retrieval running blind. A "today / this week / tomorrow" question is also flagged here, so the answer is grounded on today's transits and panchang and is forbidden from reusing the native's birth tithi.
3.5 · 5.5
An early cache shortcut, and a direct-calculation fast path for questions that are pure computation. Both return a real answer without an LLM in the loop. A cache hit is still billed at the one-cent floor rather than free, because the wallet stores integer cents and reporting a price you do not charge is its own kind of lie.
04
Yoga detection and classical-text retrieval run concurrently, filtered by the topic detected at step 3. A KP question that is chart-bound is enforced here: if the sub-lord and significator payload could not be computed, the request fails in-system with the reason rather than quietly answering from a non-KP fallback.
05
D9 (navamsa), D10 (dasamsa) and D7 (saptamsa) are derived from the rashi chart — marriage, career and progeny respectively. Which of them reaches the prompt depends on the topic, not on the model.
6.5
A deterministic pass that extracts chart patterns, looks up the classical texts for them, and composes an interpretation document. This is the step that makes the difference between a model reading a chart and a model rendering an interpretation someone else computed. Yogas and doshas go in as named, resolved facts.
07
Twenty-two sections, gated by a matrix rather than by string matching: topic, whether a partner chart is present, whether there is conversation history, and which system is active all decide what is included. Only the active system's fact family is emitted — a Western question does not receive Parashari fact blocks, and KP and Jaimini get their own. If the birth time is missing, every house, lordship, dasha, divisional, KP and yoga block is suppressed, because all of them derive from an ascendant that cannot be trusted, and a directive tells the model to say so.
08 · 8.7
The tier and timeout are chosen, and the wallet is reserved before the call, keyed on a hash of the idempotency key plus the request body — so a retry reuses the same hold instead of charging twice, and a different question under the same key is correctly a different reservation.
09
It is handed the computed chart, the derived interpretation and the retrieved classical material, and asked to write. It is not asked where a planet is. The call races a timeout; a failure past this point releases the hold rather than billing for nothing.
10 · 10.5 · 11
A validator, a hallucination gate and a scrubber run over the generated text, followed by a sentence-level surgical rewrite that corrects claims against the computed chart rather than discarding the whole answer. This is the step that catches a model that wrote a plausible placement instead of the real one.
11.5 · 12
The length cap runs before the load-bearing suffixes and the safety passes, so truncation can never remove them. Translation happens after that, and adds its own surcharge to the cost tracker rather than being folded invisibly into the query price.
12.4 · 12.5
If the answer is empty or whitespace after every pass, a chart-grounded "ask once more" line is substituted and the hold is released — the contract is never an empty body, and you are never billed for a non-answer. An independent completeness check runs alongside it.
13 · 14
Every exit path — success, refusal, cache hit, timeout recovery, the never-empty substitution — persists in exactly one place, so a conversation cannot be double-saved by construction. The request-id replay cache is written before the content cache, so a retry in the window between them cannot be charged twice.
The placements in an answer are not the model's recollection of astrology; they are the output of the same deterministic code that serves the other 558 operations. That is why the chart comes back in the response — you can diff the prose against it, and so can your support team.
Determinism in the prose. The chart is reproducible; the sentences are a model's. The pipeline constrains what the model may assert and checks it afterwards, which is a much stronger claim than "we use a good prompt" — but it is not the claim that two identical requests produce two identical paragraphs.
The request
system takes four values. Switching it
does not change the shape of the request, and this page is not going
to pretend otherwise — the honest demonstration is that the body is
identical apart from one string, and that what changes is downstream:
which fact blocks are assembled at step 7, and which grounding is
enforced at step 4.
Request body — vedic
{
"question": "What does my chart say about career?",
"birthDetails": {
"datetime": "1990-05-15T10:30:00+05:30",
"latitude": 28.6139,
"longitude": 77.2090,
"timezone": "Asia/Kolkata"
},
"system": "vedic",
"language": "en"
}
system is omitted.Request body — western
{
"question": "What does my chart say about career?",
"birthDetails": {
"datetime": "1990-05-15T10:30:00+05:30",
"latitude": 28.6139,
"longitude": 77.2090,
"timezone": "Asia/Kolkata"
},
"system": "western",
"language": "en"
}
Request body — kp
{
"question": "Will I change jobs this year?",
"birthDetails": {
"datetime": "1990-05-15T10:30:00+05:30",
"latitude": 28.6139,
"longitude": 77.2090,
"timezone": "Asia/Kolkata"
},
"system": "kp",
"language": "en"
}
datetime to the minute, and expect a validation error rather than a vague answer if you cannot.Request body — vastu
{
"question": "My kitchen is in the north-east. Is that a problem?",
"system": "vastu",
"language": "en"
}
system: "vastu", but routing is topic-driven — a vastu question reaches the vastu fact block whichever system you name, and that block is self-gated by topic.The wire format is camelCase. Every multi-word field also accepts its snake_case spelling as an alias, because that is a common SDK and agent convention and silently dropping it loses conversations.
| Field | Type | Required | Notes |
|---|---|---|---|
| question | string | Required | 1–2000 characters. Over 2000 is rejected with QUESTION_TOO_LONG before the wallet is reserved, so an oversized question costs nothing. |
| birthDetails | object | Optional | datetime, latitude, longitude, timezone. Flat top-level fields, a birthData object, and birthDate+birthTime or date+time pairs are all folded in. Timezone is inferred when it can be. |
| partnerBirthDetails | object | Optional | Same shape. Its presence is what switches on the synastry section of the prompt. |
| system | enum | Optional | vedic · western · kp · vastu. Defaults to vedic. |
| language | string | Optional | One of 30 codes. Defaults to en. |
| speed | enum | Optional | standard · fast · eco. This is the field that selects the delivery tier. |
| conversationId | string | Optional | Send back what you were given to continue a thread. Omit it and a new id is minted. conversation_id also deserialises. |
| conversationHistory | array | Optional | { role, content } messages, if you would rather carry the thread yourself than use the id. |
| systemPrompt | string | Optional | Your own persona layer. Capped at 400 lines / 12,000 characters, rejected with SYSTEM_PROMPT_TOO_LARGE rather than silently truncated. Aliases: system_prompt, customPrompt, custom_prompt, persona. |
| maxTokens | integer | Optional | Upper bound on the generated answer. |
| temperature | number | Optional | Passed through to the model. |
| responseFormat | string | Optional | The value "voice" is a server-internal signal and is stripped from any client request, so a text integration cannot accidentally opt into the spoken tone. |
| stream | boolean | Optional | Defaults to false. There is also a dedicated SSE endpoint at /api/v1/astrology/query/stream, in both GET and POST forms. |
Your own persona
systemPrompt can and cannot doThere is a real prompt surface here, and it is worth being precise about its boundary, because "you can send a system prompt" is usually sold as more than it is. Yours is a layer on top of the assembled prompt, not a replacement for it.
language field.A persona that is doing something useful
You are Tara, the assistant inside a career app. Answer in at most 120 words. Open with the single clearest signal in the chart, then one concrete suggestion the reader can act on this month. Never use the words "destiny" or "fate". Do not add a disclaimer; the app already shows one.
That text goes in the systemPrompt field
as a single JSON string, newlines escaped as
\n — printed here as real lines because
that is how you will write it.
Notice what is not in there: no chart data, no "you are an expert astrologer", no instruction about which house rules career. All of that is already assembled at step 7 from the computed chart. Your persona is for the things the pipeline cannot know — your product's voice, your length budget, your house style.
The streaming endpoint additionally accepts an Enterprise white-label persona — a custom assistant name, system prompt and sign-off — as a first-class part of its request body rather than as a prompt hack.
Rendering
This is the part most integrations get half-right, so here is the
whole contract. The response serialises camelCase — if you are reading
follow_up_suggestions or
conversation_id off the wire, you are
reading undefined.
| Field | Always present | What to do with it |
|---|---|---|
| success | Yes | Boolean. Check it before anything else; a 200 with success: false is a real state. |
| response | Yes | The prose. This is the field you render. See the whitespace note below. |
| answer | No | A legacy alias for response, present for older SDK consumers. Do not write new code against it; do not treat its absence as an error. |
| birthChart | Yes (null when there is no chart) | The differentiator. The structured chart the answer was written from. Render it beside the prose — see below. |
| followUpSuggestions | Yes ([] when there are none) |
Array of strings. Becomes your next row of chips. Always an array, so you can map it without a guard — but it can be empty, so guard the row, not the field. These are templated by topic, not generated by the model: a career question gets the career set. That makes them fast, free and predictable, and it means they are suggestions rather than insights. |
| followUps | Yes ([] when there are none) |
Byte-identical to followUpSuggestions — both fields are populated from the same array for wire parity with an older client. Read either; do not concatenate them. |
| conversationId | No | Send it back on the next turn to continue the thread. Store it against the user's session, not against the message. |
| requestId | No | Log it. It is the one string that lets support find your exact call. |
| metadata | Yes | Timing, token usage and cost for the call. Useful for your own dashboards; not something to show an end user. |
response is plain prose with real
newlines in it, including blank lines between paragraphs. That is
enforced twice: the default persona forbids headings, bullets,
tables and **bold** outright, and a
scrubber strips any Markdown that survives anyway. So render the
characters you were sent — white-space: pre-wrap,
or split on \n\n into paragraphs. Drop
it into HTML unchanged and every newline collapses, which is what
makes an otherwise correct integration look like a wall of text.
The one exception. When an enterprise customer
supplies a custom systemPrompt, the
format scrubber is skipped — so if your persona asks for structure,
you get structure, and you then need a renderer for it. Choose one
behaviour deliberately rather than discovering it in production.
birthChart exists so the reader can
check the prose. A compact strip — ascendant, moon sign, current
dasha, the named yogas — under or beside each answer turns "the AI
said so" into "here is what it was reading". It is the cheapest
trust feature in the integration and almost nobody ships it.
Keep conversationId from the first
response and send it as conversationId
on every subsequent turn. Past eight turns the pipeline generates a
rolling summary for you, deterministically and before the chart is
computed — you do not need to trim the history yourself. If you
would rather own the thread, send
conversationHistory instead.
Render followUpSuggestions as tappable
chips under the answer, and on tap send the chip's text as the next
question with the same
conversationId. That is the entire
mechanism. It is also the single highest-leverage thing you can add
to an astrology chat, because most users do not know what to ask
second.
The response, in full
{
"success": true,
"response": "…the prose, with real \n\n between paragraphs…",
"answer": "…legacy alias for response…",
"birthChart": {
"ascendant": "Pisces",
"moonSign": "Sagittarius",
"currentDasha": "Moon-Mars",
"keyYogas": ["Gajakesari Yoga", "Budha-Aditya Yoga"]
},
"followUpSuggestions": [ "…", "…", "…" ],
"followUps": [ "…" ],
"conversationId": "conv_…",
"requestId": "…",
"metadata": { engine, timing, usage, cost }
}
The birthChart keys above are the ones
the keyless sandbox returns, so you can render against them before you
have an account. Production returns the full structured chart.
What it looks like
Both frames below are CSS — no images, no canvas, no script. The text inside them is a real recorded exchange, quoted verbatim and labelled with its tier and latency; the layout is the pattern described above, with the chart strip sitting under the answer where the reader can check it.
What does my chart say about career?
Based on Vedic astrology principles, your chart shows a strong emphasis on career and communication.
What does my chart say about career?
Based on Vedic astrology principles, your chart shows a strong emphasis on career and communication.
Try it here
There is no video of this because a recording of an API call is worth
less than an API call. The button below makes an actual HTTPS request
from your browser to
https://api.vedika.io/sandbox/chat/ask,
which is mounted outside the authentication stack — no key, no wallet,
no billing. It returns static sample data in the production response
shape, and it says so in its own answer text. If the request fails you
will see the failure, not a fabricated success.
Result
Nothing sent yet. The response will be printed here exactly as it arrives.
Recorded answers
These are not written for this page. They are records pulled from a
13,740-case evaluation corpus, joined to their completions by case id,
and quoted exactly — including the formatting, the length and the
latency. Every one has finish_reason: "stop"
and passes all five of its graders. Where a completion is trimmed to
fit, the cut is on a sentence boundary and the true full length is
stated. Nothing here has been polished, corrected or shortened for
effect, and no astrological claim on this page was written by us.
Question
What does my chart say about my marriage?
Completion — first 604 of 2,610 characters, cut at a sentence boundary
Based on the provided birth data, your Vedic astrological chart indicates a complex but ultimately positive outlook for marriage. **Key Indicators:** * **The 7th House (Kalatra Bhava):** Your 7th house, which signifies the spouse and the nature of the marital union, is Capricorn (Makara). The lord of this house is Saturn (Shani). Saturn is placed in your 8th house in Aquarius (Kumbha). This placement can indicate a spouse who is mature, serious, and perhaps from a different background or age group. It can also point to potential delays or challenges in the initial stages of finding a partner.
Question
जन्म 14 मार्च 1991, सुबह 07:42 बजे, लखनऊ। KP का उपयोग करके, मेरे 7वें कस्प (cusp) का सब-लॉर्ड बताता है कि मेरी शादी कब होगी?
Completion — complete, 583 characters, untrimmed
आपके जन्म कुण्डली के अनुसार, 7वें भाव के कस्प का सब-लॉर्ड शनि है। के.पी. ज्योतिष में, विवाह का समय मुख्य रूप से 7वें भाव के संकेतकों (significators) के महादशा और अंतर्दशा से निर्धारित होता है। आपके लिए 7वें भाव के मुख्य संकेतक शुक्र (7वें भाव के स्वामी) और शनि (7वें कस्प के सब-लॉर्ड) हैं। विवाह की संभावना तब अधिक होती है जब शुक्र या शनि की महादशा या अंतर्दशा चल रही हो, या कोई अन्य ग्रह जो 7वें भाव का संकेतक हो और शनि से संबंधित हो (जैसे शनि की नक्षत्र में स्थित हो या शनि पर दृष्टि रखता हो)। यह सिद्धांत के.एस. कृष्णमूर्ति द्वारा प्रतिपादित के.पी. ज्योतिष का आधारभूत नियम है।
Question
What is my life path number and what does it mean for me?
Completion — first 609 of 1,019 characters, cut at a sentence boundary
Your Life Path Number is 9. This calculation is based on the Pythagorean system of numerology, attributed to the ancient Greek philosopher Pythagoras. The Life Path Number 9 is the number of a humanitarian, a philanthropist, and a universalist. Individuals with this Life Path are characterized by their deep compassion, selflessness, and a strong desire to help others. They are often drawn to careers or activities that serve the greater good, such as social work, teaching, healing, or activism. The core lesson for a Life Path 9 is to learn the balance between giving to others and caring for oneself.
Question
ഞാൻ 14 മാർച്ച് 1990, 04:25 AM, ബോസ്റ്റൺ എംഎയിൽ ജനിച്ചു. എന്താണ് എൻ്റെ സൺ, മൂൺ, റൈസിംഗ് സൈൻ?
Completion — complete, 249 characters, untrimmed
Pisces, Sagittarius, Capricorn. ക്ലാസിക്കൽ ജ്യോതിഷ ഗ്രന്ഥങ്ങളിൽ, പ്രത്യേകിച്ച് ടോളമിയുടെ *ടെട്രാബിബ്ലോസിൽ* (Tetrabiblos) പ്രതിപാദിച്ചിരിക്കുന്നതുപോലെ, ക്രാന്തിവൃത്ത നക്ഷത്രരാശികളാലാണ് (ecliptic constellations) ഈ രാശികൾ നിർവചിക്കപ്പെട്ടിരിക്കുന്നത്.
A note on the fourth one: it answers in three English words and then cites Ptolemy's Tetrabiblos in Malayalam. That is the method-cite behaviour working — the answer names where the rule comes from rather than asserting it — and it is also, at 249 characters, shorter than we would have written if we were writing them. We are not.
SDKs
Every signature below was read out of the package source rather than written to look idiomatic, because a snippet that does not compile against the published version is worse than no snippet. Where an SDK does not reach this endpoint, that is what it says.
@vedika-io/sdk · 3.0.6
npm i @vedika-io/sdk
// VedikaClient, from './client'
const vedika = new VedikaClient({
apiKey: process.env.VEDIKA_KEY, // required
baseUrl: "https://api.vedika.io", // optional
timeout: 60000, // optional
maxRetries: 3, // optional
cacheEnabled: true, // optional
language: "en" // optional
});
const res = await vedika.askQuestion({
question: "What does my chart say about career?",
birthDetails: { … },
system: "vedic", // 'vedic' | 'western' | 'kp'
speed: "standard", // 'standard' | 'fast' | 'eco'
conversationId, // optional
partnerBirthDetails // optional
});
askQuestion POSTs to /api/v1/astrology/query and returns the parsed body. There is also askQuestionStream, which returns an AsyncGenerator<string> over the SSE endpoint.system union lists three values and omits "kp"'s sibling "vastu", which the API does accept. If you need Vastu today, cast or call the endpoint directly — the wire accepts it.response (or the answer alias) and take timing and cost from metadata, which is where the server actually puts them.@vedika-io/react · 1.0.0
npm i @vedika-io/react @vedika-io/sdk
// wrap the tree once
<VedikaProvider apiKey={key}>…</VedikaProvider>
// then, in a component
const { data, loading, error, ask, reset } = useAskVedika();
await ask({
question: "What does my chart say about career?",
birthDetails: { … }
});
// streaming variant
const stream = useAskVedikaStream();
useAskVedika is a state wrapper — data, loading, error, ask, reset — around the JavaScript SDK's askQuestion. It has no HTTP logic of its own, so @vedika-io/sdk is a peer you install alongside it.useVoice that targets a different route.vedika-sdk · 3.0.6
pip install vedika-sdk
from vedika import VedikaClient
client = VedikaClient(
api_key=os.environ["VEDIKA_KEY"],
base_url="https://api.vedika.io",
timeout=60,
max_retries=3,
cache_enabled=True,
language="en",
)
res = client.ask_question(
question="What does my chart say about career?",
birth_details={ … },
language="en",
# keyword-only from here
system="vedic",
speed="standard",
conversation_id=None,
partner_birth_details=None,
)
language is keyword-only, which is deliberate — positional arguments on a call with nine optional parameters age badly.QuestionResponse dataclass. As with the JS package, prefer the raw response / answer text and read timing and cost from the response metadata.ask_question_stream is the SSE counterpart.vedika_sdk · 1.0.0
dependencies:
vedika_sdk: ^1.0.0
final vedika = VedikaClient(apiKey: key);
// 23 domain services — astrology, western, tarot,
// matrimony, panchang, … geocode.
await vedika.astrology.birthChart(…);
// The conversational endpoint is NOT wrapped.
// Call it directly until it is:
final res = await http.post(
Uri.parse('https://api.vedika.io/api/v1/astrology/query'),
headers: {'X-API-Key': key, 'Content-Type': 'application/json'},
body: jsonEncode({'question': q, 'birthDetails': bd}),
);
/api/v1/astrology/query. That is a gap in the SDK, not a limitation of the API.http.post is the whole integration — it is a single JSON body, and you are not missing any client-side logic by calling it directly.get() helper takes queryParams:, which is unrelated to this endpoint despite the name.Failure
Every error is JSON with an error message
and a stable code. Branch on the code,
never on the message. Some codes add fields — a 402 carries the wallet
shortfall, a 429 carries your limits and a retry delay.
| Status | Codes you should handle |
|---|---|
| 400 | VALIDATION_ERROR, QUESTION_TOO_LONG (adds maxLength and actualLength), SYSTEM_PROMPT_TOO_LARGE |
| 401 | MISSING_AUTH, INVALID_API_KEY, KEY_EXPIRED, INVALID_FORMAT |
| 402 | Insufficient balance — the body carries wallet: { required, available, deficit } and a purchase URL. Also SUBSCRIPTION_EXPIRED. |
| 403 | SUBSCRIPTION_INACTIVE, CONVERSATION_NOT_FOUND, CONVERSATION_ACCESS_DENIED, TEST_KEY_REJECTED |
| 410 | CONVERSATION_EXPIRED — mint a new thread rather than retrying. |
| 429 | RATE_LIMIT_EXCEEDED, IP_BLOCKED |
| 503 | AI_ERROR, ECO_UNAVAILABLE, CONVERSATION_STORE_UNAVAILABLE — retryable. |
| Plan | Req/min | On this endpoint |
|---|---|---|
| Unrecognised | 10 | 10 |
| Starter | 50 | 25 |
| Professional | 200 | 100 |
| Business | 500 | 250 |
| Enterprise | 1,000 | 500 |
| Flagship | 5,000 | 2,500 |
/api/v1/astrology/query carries a 0.5
multiplier on the per-minute limit, floored at 10 — it is the most
expensive call in the API and it is throttled accordingly. Burst
capacity is a tenth of the per-minute figure, minimum 2. Read
X-RateLimit-Limit,
X-RateLimit-Remaining and
X-RateLimit-Reset, which are on every
authenticated response, not just the 429s.
On a 429, honour Retry-After. On a 503
or a dropped connection, retry with the same
Idempotency-Key you sent the first
time: the reservation is keyed on that plus the body, so the retry
reuses the existing hold instead of opening a second one. Do not
generate a fresh key on retry — that is the one thing that turns a
safe retry into a double charge.
Coverage
The conversational endpoint is a front door onto a much larger
calculation surface, and sometimes you want the numbers rather than
the paragraph. Every figure below is counted from the published
OpenAPI document at
https://vedika.io/.well-known/openapi.json
— so you can fetch it and check the arithmetic yourself rather than
taking a marketing number on trust. Each operation is counted once,
under its first tag, which is why the twelve columns below add up to
exactly 559.
235
Vedic astrology — the core Jyotish surface, from divisional charts and dasha timelines to muhurta and dosha analysis.
62
AI query — this endpoint and everything around it: two streaming forms, conversation sessions, a 1,000-question batch job, usage and wallet, PDF reports, voice, transit alerts and enterprise licensing.
58
Western astrology — aspect patterns, antiscia, astrocartography lines and city activations.
26
Tarot — card lookups, deterministic card-of-the-day, astro-fusion spreads and free-form interpretation.
23
Calendar — panchang, auspicious and inauspicious windows, anandadi yoga, disha shool.
15
Chinese astrology — BaZi four pillars, lunar calendar conversion, zodiac compatibility.
14
Matchmaking — Ashtakoota guna milan, dosha cancellation, D9 marriage analysis, bulk candidate ranking.
12
Daily horoscope — the daily bundle, and a white-label feed in JSON or RSS for syndication.
10
Crystals — birth crystals, and lookups by chakra, planet or sign.
10
Human Design — the full BodyGraph: type, strategy, authority, centres, gates and channels.
10
Spirituality — ishta devata, fasting schedules keyed to the weakest planets, karma analysis.
84
Everything else across 15 further tags — health, I Ching, lifestyle, career & finance, biorhythm, dreams, angel numbers, calculators, Lenormand, palmistry, runes, widgets, oracle, voice and geocoding.
26 tags, 702 unique paths, 704 operations — two paths carry a second method, which is why the operation count runs two ahead of the path count. The conversational endpoint is one of the 704.
Vaastu
Vastu is the one tradition here that has nothing to do with a birth
chart, so it has its own surface: geometry in, classical verdict out.
Ask about it conversationally with
system: "vastu", or call these directly
when you need the polygons and the grades. Every description below is
the operation's own summary from the served spec.
Audit a full floor plan against classical zone rules, returning an overall score and grade plus a per-defect list with severity, recommended zone and remedy.
Score and defects plus a 45-devata zone heatmap, an optional mandala projection over the plot polygon, and a severity-ordered remediation walkthrough.
Check whether one room is placed in a compliant zone, returning compliance status, severity, the recommended zone and a remedy.
Correct a magnetic compass reading to a true-north zone by applying the local declination, returning the corrected zone and bearings.
The offset between magnetic and true north at a given latitude, longitude and date, with a plain-language interpretation for compass correction.
Map a bearing in degrees to its 8-zone direction label, returning the presiding deity, element and the rooms prescribed and forbidden in that zone.
Classify a main-door position into its 32-pada segment from the plot polygon and orientation, returning the pada deity, auspiciousness grade and effect.
Project the 81-cell Paramasayika Vastu Purusha Mandala onto a plot polygon, returning each cell's world-coordinate polygon, presiding devata and consequence on disturbance.
Project the 9-zone mandala onto a plot polygon, returning each directional cell's polygon, deity, element and prescribed or forbidden rooms.
Project the sacred centre of the mandala onto a plot polygon, returning the centre polygon, advisory clear-space buffer and forbidden actions.
Analyse a plot's length-to-width ratio against the classical 2:1 limit, returning the measured ratio, category, acceptability and any remedy.
Classify a plot polygon's shape — square, rectangle, L-shape, triangle, irregular, circular — and grade its auspiciousness from the fill ratio and vertex count.
Validate the quality of an AR room scan before running an audit, returning an A–F grade, missing-data blockers, warnings and re-scan suggestions.
The defect catalogue with codes and severity grades, the 8 directions with Sanskrit names and bearing ranges, the full 45-devata table, the 9-zone table, and the remedies catalogue.
Reach
Set language on the request. The list is
15 Indic plus 14 international, each backed by a bundled phrase
catalogue, plus English — which needs no translation but is a valid
value. A language is only on this list if the catalogue file exists;
it is a contract, not an aspiration. Translation runs at step 12 and
adds its own surcharge rather than being hidden in the query price.
en Englishhi Hindibn Bengalite Teluguta Tamilgu Gujaratikn Kannadaml Malayalammr Marathipa Punjabiod Odiaas Assamesene Nepalisi Sinhalaur Urdusa Sanskritfa Persianes Spanishfr Frenchde Germanpt Portugueseru Russianja Japaneseko Koreanzh Chinesear Arabicth Thaivi Vietnameseid Indonesianms MalayWhat it costs
Buying a plan credits a wallet with that month's balance; queries debit it. There is no free production tier, no trial and no credit grant — the business rule is written that way in the billing code, not just on this page. The floor on any billable call is one cent, because the wallet stores integer cents and reporting a price below what is actually debited would be a fiction.
$12 /mo
$60 /mo
$120 /mo
$240 /mo
Set by speed on the request. Which are
available to you depends on the plan.
$0.01 – $0.03
Per query. speed: "standard", and the default.
$0.02 – $0.05
Per query, on eligible plans. speed: "fast".
Plan dependent
The premium routing. The rate is set by the contract rather than published here.
$0.01
The one-cent floor. Reported and debited identically — never advertised lower than it is charged.
The two ranges above are the published rates. Underneath them the
charge is metered, not fixed: the cost of a call is computed from its
real input, output and cached token counts at the model's own rate,
multiplied by a flat markup, plus a per-query infrastructure
component, and then floored at one cent. So a short question costs
less than a long multi-turn one on the same tier, and the ranges are
the band that falls out of that rather than a price list. Every
response carries its own metadata.cost
and metadata.usage, so you never have to
estimate — you can read what the call cost from the call itself.
The keyless sandbox is not a tier: it serves static sample data with no authentication and no billing, so you can wire the integration up before you have an account. See the full pricing page.
Enterprise
Four things, all of them shipping capabilities rather than a sales page. We are not going to list service levels, discounts or support tiers here, because those are contract terms and this page only carries what is in the product.
Up to 1,000 astrology questions submitted as a single asynchronous job, on Professional and Enterprise plans. The right shape for backfilling a user base rather than looping the synchronous endpoint.
The SSE endpoint accepts an Enterprise white-label persona as part of its request body — a custom assistant name, system prompt and sign-off — so your assistant is yours in the stream, not just in your wrapper.
Send an Idempotency-Key header. The
reservation key is a hash of that key plus the request body,
so a genuine retry reuses the hold and a different question under a
reused key is correctly a separate charge. This is the difference
between an idempotency story and an idempotency guarantee.
A 400-line / 12,000-character custom persona, validated before billing and before the model call, and rejected rather than truncated when it is too large. It cannot override the safety layer or the computed facts — which is the property you want when the persona is being edited by a non-engineer.
Questions
It lets your users ask any astrology question in natural language and returns an answer grounded in a real computed chart, through one conversational endpoint: POST /api/v1/astrology/query. You send a question plus birth details and get back a written answer, the chart summary it was based on, and follow-up suggestions. Answers are available in 30 languages.
They are grounded in a real chart. Vedika computes the chart facts in code first — using the open-source XALEN Ephemeris — and then writes the answer from those facts, so the chatbot does not invent placements, dashas or yogas. The structured chart it used is returned alongside the text so you can verify it.
Users can ask across Vedic, Western and KP, and the chatbot replies in the language of the question. Responses are available in 30 languages, including 15 Indic languages with native script.
Yes. For Jyotish answers, interpretations are grounded in the classical corpus — Brihat Parashara Hora Shastra, Phaladeepika, Saravali and Jaimini — rather than generic AI training data, with an honest method-cite for systems that have no single classical source.
The conversational endpoint has a free sandbox you can call without an API key. Paid plans start at $12 per month with pay-per-use wallet credits, and per-query pricing runs roughly $0.01 to $0.05. There is no free production tier.
Call the sandbox with no key and see the response shape, then swap in a key and the same request runs against production.