AION Indian Market Intelligence (IMI)
Indian market intelligence API for caller-supplied financial and policy events. IMI adds structured context to a headline: how an event may relate to Indian market sectors and stakeholder groups. It is a hosted API product; this repository contains client and integration material, not the inference model or its weights.
Documentation map: API specification · Documentation index · Public-surface publication status. Public product/integration documents are maintained here; website and external platform copy are reviewed derivatives.
Availability, 10 November 2026: New passkey enrollment is temporarily unavailable. The public Free API's configured limits are not proof that a new key can currently be issued. Check current access terms. Published security fixes, plan-field decisions and optional processing behavior must not be described as active before release verification.
What IMI does
A polarity score can say that a headline reads positive or negative, but that single label does not show which sectors may be affected in different ways or which stakeholder perspectives matter. IMI accepts one caller supplied headline, with optional publication time and article body, and returns structured Indian market context, including a signed sector_vector, confidence, market context, sector summaries and stakeholder views.
The caller supplies the news text. The API does not ingest news feeds, fetch RBI or SEBI publications, parse PDFs or filings, or monitor markets continuously. Policy and regulatory examples are supported as caller supplied text; coverage depends on the input and is not guaranteed.
Who can use it
- AI and application developers can add returned context to an LLM prompt or their own retrieval augmented generation (RAG) records.
- SaaS and CRM teams can join sector outputs to their own customer or account classifications. IMI does not connect to or enrich a CRM by itself.
- Analysts and institutional teams can process headlines from their own sources and route selected results to a review or reporting workflow.
- MCP-compatible agents can call the hosted API through the published MCP wrapper when their plan and API key allow it.
These are integration patterns built by the consumer around IMI's response. They are not bundled feed ingestion, CRM, RAG, alerting, reporting or risk-management products.
Architecture and integration boundary
IMI is a hosted, request-response intelligence component. Your application supplies an individual headline and optional context; IMI returns a structured interpretation for downstream use. Your organization remains responsible for selecting, licensing, collecting, deduplicating and validating its source material.
flowchart LR
subgraph Customer["Your application / institution"]
A["Authorized news, policy, research or internal event source"]
B["Your collection, source verification and deduplication"]
C["Your server-side API client"]
F["Your sector mapping, review logic and evidence store"]
G["Your dashboard, RAG, CRM or analyst queue"]
A --> B --> C
F --> G
end
subgraph AION["AION hosted service"]
D["POST /v1/analyze<br/>API key and plan checks"]
E["Event context and sector interpretation<br/>Structured JSON response"]
D --> E
end
C --> D
E --> F
The boundary matters: IMI does not operate the source collector, push an event stream, host your data warehouse, resolve your customer or portfolio exposures, send notifications, or make decisions for your application. These are integration layers you build or procure separately.
From headline to consumer-facing context
flowchart TD
A["One caller-supplied headline<br/>optional published_at and body (REST)"]
B["Hosted IMI analysis"]
C["Signed sector_vector and contextual fields<br/>subject to response and plan contract"]
D{"Does the output contain<br/>usable supported context?"}
E["Store as incomplete / unsupported<br/>route for review"]
F["Associate with your own event ID,<br/>original source and sector mapping"]
G["Human-reviewed research, alerts<br/>or downstream enrichment"]
A --> B --> C --> D
D -->|No / missing / uncertain| E
D -->|Yes| F --> G
The diagrams describe consumer integration patterns, not a disclosure of private model architecture, a guaranteed inference result, or an implemented AION queue/alert service.
Practical integration patterns
LLM or RAG enrichment
Store the returned sector_vector and stakeholder context with the original item in your own retrieval system. When that item is retrieved, your application can supply the structured context to its LLM. IMI does not provide the vector database or retrieval layer.
Policy and institutional monitoring
Send a headline and, where available, its article body from your own RBI, SEBI, government, company or news source. Your application can review the returned sector and stakeholder context alongside the original source. The API does not fetch or authenticate those publications.
Sector aware SaaS and CRM workflows
Your application may compare returned sector values with its own customer or supplier classifications to prioritize human review. Any thresholds, account mapping and resulting actions are your application's logic; IMI does not supply a CRM connector or make business decisions.
Research queues and alerts
Process your own headlines one request at a time, then sort or route results using the returned fields. The API does not provide a feed, batch job, scheduled alert service or research report.
MCP agent workflows
The MCP package exposes analyze_news over the same hosted API. It is an API wrapper for MCP clients, not a model running locally or inside Hugging Face.
Integration patterns by industry
These are examples of customer-built applications, not bundled IMI products or claims about current customers.
| User or industry | Event supplied by your system | How to use IMI's response | What remains your responsibility |
|---|
| Fintech research platforms | An authorized RBI, SEBI or company headline | Present sector-level context beside the original item; let an analyst compare alternatives | Source rights, timestamps, evidence and editorial review |
| Portfolio and exposure analytics | A macro or policy event | Join the sector_vector to your own sector-to-holdings mapping and flag exposures for review | Holdings data, mapping, suitability, risk calculations and decisions |
| Bank/NBFC research teams | A funding, monetary-policy or regulatory event | Attach structured sector context to an analyst work item | Credit assessment, regulatory interpretation and approvals |
| Brokerage and developer tools | A customer-licensed news event | Enrich a research screen or internal event queue with classification context | Broker integration, market data rights, alerts and execution controls |
| Compliance and public-policy teams | A government or regulator statement supplied as text | Triage items for human assessment of potentially affected sectors | Authenticity, legal interpretation, obligations and filing review |
| Enterprise procurement and supply chains | A commodity, weather or economic headline | Compare affected sectors with the company's own supplier categories | Supplier ontology, exposure mapping and business action |
| SaaS / CRM vendors | A customer-provided news item | Annotate client segments or work queues using a verified mapping | CRM connector, account data and workflow automation |
| LLM, RAG and agent builders | A retrieved headline with source metadata | Attach structured sector and stakeholder context to the original source | Retrieval, provenance, citations, prompt construction and hallucination checks |
| Education and research | A dated macro-event headline | Use a reviewable structured case for comparison or teaching | Methodology, source review and assessment of uncertainty |
Do not turn an IMI sector score into an automated trade, credit decision, compliance determination or other consequential action without independent controls and human review. A signed contextual value is not a forecast of price movement or a probability of economic loss.
The managed endpoint accepts one headline, optional published_at, and optional body per request. The article body can add context when the headline is ambiguous. It does not accept a feed, PDF, filing, table or multi-event batch through this endpoint.
The documented public response includes:
headline — the submitted headline.
confidence — the system's classification confidence; it is not a probability of a market outcome.
macro_signal and vix_regime — market-context fields when returned.
sector_vector — signed sector values; this is the canonical sector-impact field.
top_positive_sectors and top_negative_sectors — compact summaries when returned.
stakeholder_views — structured stakeholder perspectives when returned.
cultural_context — a plain-language indicator and explanation when a supported cultural or sovereign context applies.
- Internal assignment identifiers and execution-oriented fields are removed from the public response. Ignore unrecognized fields rather than depending on them.
API keys are model and plan scoped and require a verified email and server-verified passkey. The free API plan is available once per verified user/account and includes JSON REST access with sector_vector and stakeholder_views: 10 successful requests per rolling 30 days, with a maximum of 10 in any rolling 24 hours. Free key issuance is also limited by device and network. Other plan quotas and integration entitlements differ; consult the current API catalogue for current terms.
The website demo key is a separate browser-demo allowance, not an API key or a second free API plan. It is issued once per email and allows up to 3 IMI analyses per browser/device pass and 30 per network in a rolling 7 days. An optional passkey raises the demo allowance to 10 analyses per passkey. Passkey verification happens on the visitor's device; AION does not receive biometric data. The demo key does not authorize calls to the production API. See the website demo access page for current access terms.
How to interpret and consume the response
The canonical sector-impact field is sector_vector, a mapping of sector names to signed numeric values. The configured public contract describes a 32-sector representation; do not assume every response has a complete, non-null or equally informative vector. Treat absent or unsupported context explicitly.
| Field | Consumer interpretation | Safe handling |
|---|
headline | Input headline as returned by the API | Associate with your own source ID; an echo is not source verification |
sector_vector | Signed relative sector-context values | Match names using your own reviewed sector mapping; do not assume a return forecast |
confidence | Classification confidence | Do not interpret as price-move probability or calibrated certainty |
top_positive_sectors / top_negative_sectors | Compact summaries, when returned | Treat as summaries of an interpretation, not trading recommendations |
stakeholder_views | Context for different stakeholder perspectives, when returned | Retain qualifiers; do not substitute for legal, financial or credit review |
macro_signal / vix_regime | Conditional market-context fields | Handle absent/null values; do not assume live quote or VIX-feed delivery |
cultural_context | Conditional context indicator and explanation | Handle missing or unsupported cases without inventing an explanation |
The exact fields and integration channels can depend on the plan and deployed contract. For the authoritative request/response shape, see API specification and the current API catalogue. In particular, do not assume that the MCP wrapper returns the full REST response.
A schema-only example, not an inference result
{
"headline": "Illustrative caller-supplied event",
"confidence": 0.65,
"sector_vector": {
"example_sector_a": 0.25,
"example_sector_b": -0.15
},
"vix_regime": null
}
All names and numbers in this block are synthetic placeholders. They are not official sector taxonomy names, a sample from the live model, measured confidence, or an expected response to a particular headline.
Integration options
REST API
POST https://api.aionanalytics.in/v1/analyze
X-API-Key: <your API key>
Content-Type: application/json
{
"headline": "RBI raises the policy repo rate",
"published_at": "2026-06-05T09:00:00+05:30",
"body": "Optional source text supplied by your application."
}
published_at and body are optional. Keep API keys on your server; do not embed them in a browser or public Space.
Python client
Install the proprietary hosted API client:
pip install aion-indian-market-intelligence
export AION_API_KEY="<your API key>"
from aion_indian_market_intelligence import analyze
result = analyze(
"RBI raises the policy repo rate",
published_at="2026-06-05T09:00:00+05:30",
)
print(result["sector_vector"])
The Python package sends requests to the managed API. It does not run inference locally.
Client version 1.0.11: The published Python analyze() interface accepts headline and optional published_at, but not body. To submit optional article-body text, use the documented REST API directly when supported by your entitlement. Do not pass body= to the Python client until a compatible client version is released.
MCP wrapper
uvx aion-indian-market-intelligence-mcp
Configure AION_API_KEY in the MCP host's server-side environment. The MCP wrapper calls the same hosted API through the Python client; REST and Python clients expose the documented public response. MCP client behavior and available plan entitlements are described in the MCP package documentation.
Choose the integration surface
| Interface | Inputs in published version | Output | Use it when |
|---|
REST POST /v1/analyze | headline, optional published_at and body | Documented JSON fields when returned and entitled | You need article-body context or direct server-side integration |
Python client 1.0.11 | headline and optional published_at; no body= argument | API response as a Python dictionary | You want a small application wrapper and have eligible API access |
MCP wrapper 1.0.6 | analyze_news(headline, published_at?) | sector_vector only | An MCP host or agent needs sector-context lookup |
Published Python client 1.0.11 host note: Its default base URL still points to a legacy hostname. Set the documented canonical base explicitly until a new client release is verified:
from aion_indian_market_intelligence import AIONClient
client = AIONClient(base_url="https://api.aionanalytics.in")
result = client.analyze("RBI announces a change in the repo rate")
sector_values = result.get("sector_vector") or {}
This is a usage example, not a record of an authenticated live request. The API requires an entitled key, normally supplied through AION_API_KEY in the server environment.
Event-driven or near-real-time integration
You can process events soon after your own event source produces them. This does not make IMI a streaming feed, and it does not establish a specific latency SLA. Each event still requires an individual, authorized API call and counts according to your current quota contract.
sequenceDiagram
autonumber
participant S as Your licensed source
participant Q as Your event queue / worker
participant I as AION IMI REST API
participant M as Your mapping and evidence store
participant H as Human reviewer / application
S->>Q: Event + source ID + publication time
Q->>Q: Verify source rights and deduplicate
Q->>I: POST one headline (+ optional time/body)
alt Authorized and interpretable response
I-->>Q: JSON sector / context fields
Q->>M: Associate result with original source
M->>H: Enriched event for review
else Authentication, quota, timeout or uncertain result
I-->>Q: Error / limited context
Q->>M: Record outcome and safe retry/review state
M->>H: Incomplete or unavailable, not zero impact
end
Minimal server-side event handler (REST)
This sample illustrates one event delivered by your own queue, webhook or scheduler. It does not create a news feed, message broker, alert dispatcher, data store or continuous IMI subscription. It makes no request unless your application invokes enrich_event.
import os
import requests
IMI_URL = "https://api.aionanalytics.in/v1/analyze"
def enrich_event(event: dict) -> dict:
"""Run once for a source event; caller owns deduplication and retries."""
payload = {"headline": event["headline"]}
if event.get("published_at"):
payload["published_at"] = event["published_at"]
if event.get("body"):
payload["body"] = event["body"]
response = requests.post(
IMI_URL,
headers={
"X-API-Key": os.environ["AION_API_KEY"],
"Content-Type": "application/json",
},
json=payload,
timeout=(5, 30),
)
if response.status_code == 429:
raise RuntimeError("IMI rate or quota limit reached; defer this event")
response.raise_for_status()
result = response.json()
if not isinstance(result, dict):
raise ValueError("Unexpected IMI response type")
vector = result.get("sector_vector")
if not isinstance(vector, dict):
vector = {}
return {
"source_event_id": event.get("id"),
"source_url": event.get("source_url"),
"sector_vector": vector,
"confidence": result.get("confidence"),
"stakeholder_views": result.get("stakeholder_views"),
"review_required": True,
"raw_context": result,
}
Your worker can then attach the returned object to a case, RAG item, internal dashboard or human-review queue. Keep the source, as-of time, API response and your own review outcome distinguishable. Do not assume downstream automation is provided by AION.
Operational design for a production consumer
- Source and custody: Obtain authorized text from your licensed/public source; preserve its source ID, URL, publication time and permission to process it. IMI does not authenticate that source.
- Deduplication: Deduplicate before calling the API using your own stable event identifiers and a clear policy for corrected or updated headlines.
- Access and quotas: Store
AION_API_KEY server-side. Model-scoped access and quotas are enforced by your entitlement; the free API tier is for limited evaluation, not an unlimited event stream.
- Bounded errors: Handle authentication/entitlement errors separately from 429 quota limits and transient network/5xx failures. Do not implement an unbounded retry storm; retries may have usage consequences.
- Partial context: Missing sectors, null regime values or unsupported inputs must route to an explicit incomplete state, never silently to a neutral/zero verdict.
- Consumer-side business logic: Any watchlist mapping, risk thresholds, notification rules, CRM action, RAG storage or human decision gate is your implementation.
- Privacy and traceability: Minimize submitted personal or confidential content. Maintain your own source/response linkage and comply with applicable retention and data-use requirements.
- Validation: Measure your own end-to-end latency, event coverage, error rate and cost against your workload. No published IMI p95/p99 or high-volume streaming guarantee is implied.
Example consumer architecture: portfolio or institutional research
flowchart LR
A["Authorized headline source"] --> B["Your IMI request worker"]
B --> C["IMI structured response"]
C --> D["Your sector-to-exposure map"]
E["Your permitted holdings / company data"] --> D
D --> F["Your exposure review screen"]
F --> G["Analyst decision and audit trail"]
A portfolio weighting, issuer classification or alert threshold is not an IMI field or an IMI-native portfolio engine. The diagram shows how a developer can join AION's sector context to its own information.
Limitations and access
- IMI interprets text supplied by the caller. It does not provide source feeds or verify the submitted source.
- One request analyzes one headline and optional body. Feed ingestion, continuous monitoring, PDF or filing extraction, and batch processing belong to your application.
- Event coverage and classification can be incomplete or uncertain. Sparse, ambiguous or unsupported inputs may produce limited context or no useful sector signal.
- Impact values are indicative context, not statistically calibrated forecasts. Confidence does not guarantee correctness or predict market movement.
- The product is information infrastructure, not investment advice, a trading signal, a buy/sell/hold recommendation, or an execution service. It does not connect to brokers or place orders.
- Hosted API access requires an API key and is quota and plan controlled. Current plan requirements and limits are listed at aionanalytics.in/apis.
- The hosted model and its weights are proprietary and are not distributed. The Python client is proprietary hosted API software. The MCP wrapper has its own license; see its repository for the current license terms.
Public documentation and distribution contract
This GitHub README is the canonical public integration narrative for IMI. It is intended to guide consistent documentation across the website's IMI model and developer pages, GitLab, Hugging Face and package descriptions.
When adapting this guide:
- Preserve the distinction between caller-supplied events and AION-provided analysis. Do not attribute source collection, streaming, batch processing or alerts to the hosted endpoint.
- Keep REST, Python client and MCP signatures and response scopes distinct. Published package behavior takes precedence over future, unpublished source changes.
- Copy diagrams only with the accompanying integration-boundary and limitations text. A customer-built workflow diagram is not evidence that AION operates the workflow.
- Show measured examples only when they come from a documented and reviewable response; otherwise label values and workflows synthetic.
- Check entitlement, demo allowance and pricing against the live API catalogue before publishing. The demo is not a production API entitlement.
- Do not publish security fixes, new identity flows, field gating or performance claims as live until their deployment and readback are verified.
- Keep model weights, private taxonomy, coefficients, training and inference implementation outside public documentation.
Links
Licensing
The hosted IMI model and Python client are proprietary. No model weights are distributed. The MCP wrapper and utility packages have separate licenses; check the license file in each repository before reuse.