A comprehensive product report on how AI agents can trade, manage money, and protect portfolios in India's derivatives market — without hallucinating a single order.
In March 2026, Public.com — a U.S. brokerage serving over 3 million users — became the world's first platform to offer AI agents for investing. Within weeks, early adopters were using natural-language prompts to automate tax-loss harvesting, rebalance portfolios, and execute complex multi-leg option strategies. The financial services industry took notice: Goldman Sachs research estimated that agentic trading could capture 15-25% of retail brokerage AUM by 2030, driven by the simple proposition that machines should handle the repetitive, rule-based parts of investing while humans focus on strategy and risk management.
For Indian brokerages, this represents both an existential threat and an unprecedented opportunity. India's Futures & Options market — the world's largest by volume — trades over ₹200 lakh crore per month. Yet every Indian broker still operates on a model that hasn't fundamentally changed in two decades: manual order placement, basic price alerts, and static watchlists. The technology stack is mature, the regulatory framework is evolving, and the data infrastructure is in place. What's missing is the intelligence layer.
This report presents a complete, end-to-end architectural blueprint for building agentic capabilities into SAHI's existing brokerage platform. It is not a theoretical exercise. Every component described here maps to SAHI's current infrastructure, existing data feeds, and regulatory obligations under SEBI. The architecture is designed to be built in phases—starting with zero regulatory risk (enhanced alerts) and scaling to fully autonomous agents over 18 months.
The core thesis is simple: SAHI already has the hardest parts — the broker license, the order management system, the real-time market data, and critically, proprietary derivatives intelligence (OI analysis, PCR tracking, Max Pain calculation, 20+ live scanners) that no other Indian platform possesses. The agentic layer — the AI that converts human intent into deterministic workflows — is the fastest and cheapest part to build. And it creates a competitive moat that is nearly impossible to replicate.
The first and most important distinction in agentic brokerage architecture is between AI-as-brain and AI-as-translator. Most people, when they hear "AI trading agent," imagine a large language model making real-time trading decisions — analyzing charts, reading news, and deciding to buy or sell. This is the wrong mental model, and it is exactly the approach that would guarantee catastrophic failures in production.
Large language models are probabilistic text generators. They produce plausible-sounding outputs based on statistical patterns in training data. This is enormously useful for understanding human intent — parsing a sentence like "sell a weekly ATM straddle on NIFTY if implied volatility exceeds 15" and extracting structured parameters. But it is categorically unsuitable for financial execution, where the same input must always produce the same output, where a rounding error can cost lakhs of rupees, and where "creative" interpretation of market conditions is the definition of unacceptable risk.
The architecture described in this report follows a fundamentally different paradigm: AI is used for setup, never for execution. During the setup phase (typically lasting 30-60 seconds), the AI is creative, conversational, and flexible. It understands complex financial intent, asks clarifying questions, and constructs a precise, reviewable workflow. But once the user approves that workflow and the agent goes live, the AI is completely removed from the execution loop. What remains is a pure rule engine — a state machine that evaluates deterministic conditions on every market tick and places orders through SAHI's existing broker API. There are no LLM calls during execution. There are no "creative" interpretations. The same market conditions will always produce the same action. This is the fundamental guarantee that makes agentic trading safe enough for real money.
User says "Buy tech stocks"
→ LLM decides which stocks to buy
→ LLM decides how many shares
→ LLM decides when to execute
→ LLM decides stop loss levels
→ Hallucination risk at EVERY step
Every LLM call in the execution loop = a potential hallucination that loses real money. LLMs can invent stock tickers, miscalculate position sizes, or "creatively" interpret stop losses.
User says "Buy tech stocks"
→ AI asks: which stocks? budget? trigger?
→ AI builds a deterministic workflow
→ User reviews every parameter
→ User approves with backtest data
→ Rule engine executes — ZERO LLM calls
Same conditions = same action. Every time. No hallucinations. Full audit trail. SEBI-compliant transparency.
Why does this distinction matter so much for SAHI? India's regulatory environment under SEBI is significantly more stringent than the US market where Public.com operates. SEBI's algorithmic trading framework (Circular SEBI/HO/MRD/DP/CIR/P/2016/30 and subsequent amendments) requires that any system executing trades automatically must have risk management controls, audit trails, and real-time monitoring capabilities. By designing the agentic architecture as a deterministic workflow system rather than an AI decision-maker, SAHI can potentially classify Phase 1 agents as "enhanced alerting systems" rather than algorithmic trading systems — significantly reducing the initial regulatory burden and allowing faster time-to-market.
Furthermore, the deterministic design addresses SEBI's specific concerns about AI in financial services. In its 2024 consultation paper on AI/ML in securities markets, SEBI explicitly flagged the risk of "opaque AI-driven execution" and mandated that any AI-assisted trading system must provide explainability for every decision. In SAHI's architecture, every agent action is fully explainable because the logic is a simple IF-THEN rule tree that was reviewed and approved by the user before activation. There are no neural network weights to interrogate, no hidden layers to audit — just straightforward conditional logic.
The AI converts natural-language intent into a structured, reviewable workflow. It does NOT directly execute trades. The workflow is a transparent JSON document that the user can read, modify, and approve before any market interaction occurs.
During setup: creative AI. During execution: ZERO creativity. Pure rule evaluation on every tick. The execution engine is a state machine, not a neural network. Same inputs always produce the same outputs — the fundamental requirement for financial-grade systems.
Agents cannot access the internet, call external APIs, or modify their own rules. They operate in a fully sandboxed environment with hard-coded boundaries. Every action is logged to an immutable audit trail. A hardware kill switch can terminate all agents within 100ms.
The architecture of an agentic brokerage is fundamentally a pipeline problem. Natural language goes in one end; deterministic market orders come out the other. The challenge is ensuring that the transformation from human intent to machine execution is lossless, auditable, and safe. This requires a minimum of five distinct processing layers, each with a specific responsibility and a clear boundary of what it can and cannot do.
The critical design decision is where to draw the line between AI and deterministic logic. In SAHI's architecture, this line is drawn after Layer 3. Layers 1-2 use large language models for natural language understanding and multi-turn dialogue. Layer 3 is the bridge — the AI constructs the workflow, but the workflow itself is a deterministic JSON document. Layers 4-5 are entirely AI-free. The human reviews the workflow (Layer 4), and a pure rule engine executes it (Layer 5). This means that 60% of the pipeline is deterministic, and the AI is limited to the understanding phase where hallucinations can be caught and corrected through the clarification dialogue.
Click any layer below to see the detailed technical specification, including the data formats, API contracts, and latency requirements for each stage of the pipeline.
User types in plain English — "If PCR drops below 0.7 and RSI < 30, buy NIFTY CE"
Multi-turn AI dialogue to eliminate ambiguity before any action
Structured JSON: Trigger → Condition → Action → Guardrails
User approves exact workflow with backtest simulation data
Zero LLM — pure rule engine evaluating conditions on every tick
The clarification loop is the single most important safety mechanism in the entire architecture. It is the point where ambiguity is resolved, parameters are confirmed, and the user's true intent is captured with precision. In the Indian F&O context, this is especially critical because of the complexity of derivatives instruments — a user who says "buy NIFTY calls" could mean weekly or monthly expiry, ATM or OTM strike, 1 lot or 10 lots, with or without a hedging leg. The clarification engine must resolve all of these dimensions before any workflow is generated. The demo below shows this process in action for a typical F&O agent setup.
Every workflow generated by the system must contain exactly four components. This is not a suggestion — it is a hard architectural constraint enforced at the schema validation layer. A workflow missing any of these four components will be rejected by the system and cannot be activated. This rigidity is intentional: it ensures that every agent has explicit triggers, conditions, actions, and safety boundaries before it can interact with real money.
Agents are organized into five categories based on their primary function: F&O Execution, Market Monitoring, Money Management, Risk Management, and Fundamental Intelligence. This categorization is not arbitrary — it maps directly to the five core activities that a retail or HNI trader performs on a daily basis in Indian markets. Each category has distinct data requirements, latency constraints, and regulatory implications.
The F&O Execution agents represent SAHI's deepest competitive moat. No other Indian brokerage — not Zerodha, not Angel One, not Groww — has the combination of real-time OI data, PCR analysis, Max Pain calculations, and 20+ proprietary scanners running natively on their platform. These data feeds become the trigger conditions for agents that are literally impossible to build on any competing platform. A user cannot ask Zerodha's Sentinel to "buy NIFTY CE when PCR drops below 0.7" because Zerodha does not compute PCR. SAHI does. This is the moat.
Each agent card below is interactive — click any card to reveal the exact deterministic trigger logic that would power that agent in production. This trigger logic is the JSON workflow that the NLP parser generates and the rule engine evaluates. Notice how each trigger uses SAHI's proprietary data feeds that are unavailable on competing platforms.
India's F&O market trades ₹200+ lakh crore/month. These agents leverage SAHI's proprietary derivatives intelligence that no competing platform can replicate.
The F&O execution category is where SAHI's data advantage becomes most apparent. India's derivatives market has unique characteristics — expiry-day dynamics, OI-driven price action, PCR as a sentiment gauge, Max Pain as a gravitational level — that create trading opportunities visible only to platforms with real-time derivatives analytics. Each agent below uses at least one SAHI-proprietary data feed as its primary trigger, making it fundamentally impossible to replicate on any platform that lacks native OI/PCR/Max Pain computation.
Sell a weekly ATM straddle on NIFTY every Thursday if IV > 15. This exploits the well-documented theta decay acceleration in the final 2 trading days before weekly expiry — a pattern unique to Indian markets where weekly expiry volume exceeds monthly.
Monitor real-time OI changes across the NIFTY options chain. When a specific strike sees massive OI accumulation (>2L in 30 min), it signals institutional activity. Buy that strike before the move completes.
The Put-Call Ratio is India's most reliable sentiment gauge. When PCR drops below 0.7 (extreme bearishness) AND RSI confirms oversold, it historically signals a reversal — a signal that no other Indian broker can automate.
Max Pain theory suggests NIFTY gravitates toward the strike where option writers' losses are minimized. When spot is >200 points away from Max Pain on expiry day, the convergence trade has a historically high win rate on SAHI's backtests.
Expiry days in Indian markets create unique gamma dynamics — large moves happen in short bursts. This agent waits for a 100+ point NIFTY move in 15 minutes and enters a straddle to capture the expansion.
▊
24/7 surveillance across price, volume, OI, institutional flows, and macro events — automating the analyst's watchlist.
Market monitoring agents solve the attention economy problem in trading. A typical active trader watches 20-50 stocks, tracks 3-4 indices, monitors sectoral moves, and follows FII/DII flow data daily. This is unsustainable human attention overhead. Monitoring agents decompose this watchlist into discrete conditions and evaluate them automatically, alerting or acting only when specific thresholds are breached. The key insight is that 90% of a trader's time is spent watching markets where nothing actionable is happening — agents eliminate this dead time entirely.
Scan all stocks in a custom watchlist for 52-week high breakouts confirmed by 2x average volume — the most reliable momentum signal in Indian markets.
FII flows are the single most correlated leading indicator for NIFTY direction. When FII turns net buyer after consecutive selling days, it historically precedes 1-3% rallies.
India VIX above 20 has historically preceded 2-5% drawdowns. This defensive agent automatically reduces F&O exposure by 50% when fear spikes, protecting the portfolio from tail risk events.
Monetary policy decisions directly impact banking, real estate, and bond markets. This event-driven agent positions the portfolio automatically on RBI MPC outcomes.
Automated cash sweeps, tax-loss harvesting, and portfolio optimization — the "set and forget" financial maintenance layer.
Money management agents handle the administrative overhead of portfolio maintenance that most investors neglect. Cash sitting idle in a brokerage account earns zero returns. Unrealized tax losses expire at year-end unused. SIP schedules require manual tracking. These agents automate the financial housekeeping that compounds into significant returns over years — Public.com's data shows that users with active cash sweep and tax harvesting agents outperform manual users by 1.2-1.8% annually, purely from elimination of behavioral inefficiency.
Any SAHI balance exceeding ₹1L is automatically invested in liquid ETFs (LIQUIDBEES/LIQUIDCASE), earning overnight returns instead of sitting idle at zero yield.
Before March 31, automatically identify positions with >₹1L in unrealized short-term losses, sell them to book the loss for tax offset, then re-enter after the wash sale period.
When market crashes >3% in a week, automatically double the SIP amount for that month — a systematic buy-the-dip strategy that eliminates emotional hesitation.
The defensive layer — automated hedging, concentration monitoring, and drawdown protection that runs 24/7.
When portfolio drawdown exceeds 5% from peak, automatically purchase NIFTY puts worth 20% of portfolio value as insurance — the institutional hedging practice made available to retail.
Prevents over-concentration in any single stock. If any position exceeds 15% of portfolio value, automatically trims it to 10% and alerts the user.
This is SAHI's deepest and hardest-to-replicate moat. The Funda Scanner integration provides access to 15 years of financial data across 2,200+ Indian companies — Balance Sheet, P&L, Cash Flow, and 50+ computed ratios (ROCE, ROE, Debt-to-Equity, Free Cash Flow Yield, Piotroski F-Score). No other Indian brokerage has a fundamental data backend of this depth. This enables a category of agents that are literally impossible on any competing platform: agents that trigger on fundamental value, not just price or technicals. A user can say "Alert me when any company with ROCE > 20%, zero debt, and PE < 15 drops 10% in a week" — and the agent will scan all 2,200 companies daily to find that exact combination. This is category-defining capability.
Continuously scan all 2,200+ companies for high-quality value opportunities: ROCE > 20%, Debt-free, PE < 15, with a recent price drop indicating market mispricing.
Monitor holdings for deteriorating fundamentals. If a stock's PE drops below 5 but ROE is also below 5%, it may be a value trap rather than a genuine value opportunity.
To make the architecture tangible, this section walks through the complete lifecycle of a real agent — from the moment a user types a natural-language prompt, through the clarification dialogue, workflow generation, user approval with backtest data, and finally the 24/7 autonomous execution with rolling rules and assignment handling. This example uses a covered call income agent — one of the most requested strategies in Indian markets — where a user wants to generate monthly premium income from their existing stock holdings.
The covered call strategy is particularly well-suited to the agentic model because it requires weekly mechanical execution (selling calls, monitoring deltas, rolling positions) that is tedious for humans but trivial for a rule engine. In the traditional manual approach, a trader must log in every Friday, check their holdings, find appropriate OTM calls, calculate position sizes, and place orders — a 20-minute process that most people forget or skip. An agent handles this in milliseconds, every week, without fail.
"Help me generate ₹50,000/month in covered call premiums on my RELIANCE and HDFC holdings. I have 500 shares of each. Use 30-delta calls, monthly expiry. Roll if losing."
The clarification engine asks: (1) Preferred delta? → 0.30 confirmed. (2) Weekly or monthly expiry? → Monthly. (3) What if shares get called away — rebuy immediately or wait for a dip? → Rebuy after 1 day. (4) Roll policy — when to roll losing positions? → Roll at 7 DTE if losing.
The system outputs a complete JSON workflow: EVERY 1st FRIDAY of month → FOR EACH [RELIANCE, HDFC] → IF owns ≥ lot_size → FIND monthly call at delta ~0.30 → CALCULATE premium → IF projected monthly income ≥ ₹50K target → SELL CALL(quantity matching shares) → LOG entry with premium collected, strike, expiry.
The workflow includes automatic adjustment logic: DAILY CHECK → IF short call at ≥ 50% profit → BUY TO CLOSE (harvest profit early) → IF position losing AND days_to_expiry < 7 → ROLL to next monthly expiry at same delta → IF underlying moves against by > 5% → ALERT user for manual review.
If shares are called away (assigned): WAIT 1 trading day → REBUY at market open VWAP → Resume covered call selling next cycle. This ensures continuous income generation without requiring user intervention on assignments. All transactions logged to audit trail with exact fill prices.
User sees: (a) Plain-English summary of the complete agent behavior, (b) The full deterministic logic tree — no hidden rules, (c) Backtest: "Based on last 6 months of RELIANCE + HDFC pricing, this agent would have generated ~₹48,200/month in net premiums with a max drawdown of ₹12,500." (d) Risk disclosure per SEBI guidelines. One tap to activate.
Building the AI agent layer is actually the easiest part of building an agentic brokerage. The hard part is the infrastructure underneath — the broker license, the order management system, the real-time data feeds, the multi-asset execution capability. These foundations take years to build and require significant regulatory capital. The strategic question for any brokerage considering the agentic model is: how many of these 7 prerequisites do we already have?
SAHI's position is unusually strong. With an active SEBI registration, a proprietary OMS, real-time NSE/BSE data feeds, and the richest derivatives analytics suite in Indian retail broking, SAHI already has 4 of the 7 foundational requirements fully operational. The three gaps — a public trading API, an NLP intent parser, and a deterministic workflow engine — are all software-only components that can be built without regulatory approvals, new licenses, or additional exchange memberships. This is a 12-month development project, not a multi-year infrastructure buildout.
Legal authority to execute trades on behalf of clients. Requires SEBI registration as a stockbroker under the Securities Contracts (Regulation) Act, 1956. Includes exchange memberships (NSE, BSE), clearing corporation membership, and compliance infrastructure.
SAHI is SEBI registered — NSE + BSE member with full clearing ✓
Complete order lifecycle management — placement, modification, cancellation, position tracking, margin calculation, and settlement. Must handle all order types: Market, Limit, Stop-Loss, GTT, GTC, After-Market Orders.
SAHI has proprietary in-house OMS with full order type support ✓
24/7 streaming market data via WebSocket — tick-by-tick prices, volumes, OI across all segments. Must include computed indicators (RSI, MACD, EMA, VIX, PCR) and derived analytics (OI analysis, Max Pain, sector rotation).
NSE/BSE live feeds + WebSocket + 100+ indicators + 20+ scanners ✓
Cross-asset execution capability on a single platform: Equity Cash, F&O (Index + Stock), Currency Derivatives, Commodity, Mutual Funds, ETFs, SGB, IPO.
Equity Cash + F&O available. MF/SGB partial. Commodity pending.
Programmatic trading interface that agents can call to place orders, query positions, fetch market data, and manage portfolios. Must be low-latency (<50ms), rate-limited, and authenticated via OAuth2/API keys.
No public API currently — CRITICAL GAP that must be addressed ✗
Large language model fine-tuned for Indian financial vocabulary — understanding F&O instruments, derivatives strategies, and multi-condition trading logic expressed in natural language.
Needs build — can leverage SAHI AI research module as foundation
Real-time rule evaluation engine that processes market ticks against agent workflow definitions, evaluates multi-condition triggers, and executes actions when all conditions are met. Must operate at <5ms latency with full audit trail.
Needs build from scratch — the core new development workstream
SAHI already has 4 of 7 foundations in place. What remains is the AI layer + workflow engine — the fastest part to build, requiring no new licenses, no regulatory approvals, and no exchange memberships.
The competitive advantage in agentic brokerage does not come from the AI — natural language processing and LLM capabilities are rapidly commoditizing. The moat comes from the data. Specifically, it comes from the proprietary data feeds that agents can use as triggers. Any brokerage can integrate GPT-4 or Gemini for intent parsing. But not every brokerage has the underlying data infrastructure that makes agents actually useful for trading. This is where SAHI's position is extraordinarily strong.
Consider the comparison table below. A standard Indian brokerage — Zerodha, Angel One, Groww — can offer agents with price-based triggers ("alert me if TCS drops 5%") because all brokers have price data. But SAHI can offer agents with PCR-based triggers, OI-based triggers, Max Pain convergence triggers, FII/DII flow triggers, and fundamental score triggers. These are signals that require proprietary analytics backends that competitors do not have. When a SAHI user says "buy NIFTY CE when PCR crosses below 0.7," the system can evaluate this condition in real-time because SAHI computes PCR natively. A Zerodha user cannot create this agent because Zerodha does not compute PCR.
This data-driven moat extends to 10 trigger categories where SAHI has exclusive capability. Of these, 7 are completely unavailable on any competing Indian platform — they show a "—" in the Standard Broker column below, meaning no existing broker can evaluate these conditions even manually, let alone through an automated agent.
Indian compliance is the single largest risk factor in building an agentic brokerage. Unlike the US where Public.com operates under SEC/FINRA rules that have relatively well-defined carve-outs for automated trading systems, India's SEBI regulatory framework for algorithmic and AI-assisted trading is still evolving. This creates both risk and opportunity — risk because regulatory uncertainty can delay launches, and opportunity because the first mover who gets the compliance framework right will have a significant defensive moat.
SEBI's current regulatory framework for automated trading is primarily governed by SEBI Circular SEBI/HO/MRD/DP/CIR/P/2016/30 (Broad Guidelines on Algorithmic Trading) and its subsequent amendments. The key distinction that enables SAHI's phased approach is SEBI's definition of "algorithmic trading" — defined as "any order that is generated using automated execution logic". Crucially, SEBI distinguishes between alerting systems (which notify users of conditions) and execution systems (which place orders automatically). Phase 1 of SAHI's roadmap deliberately operates in the alerting category, which carries zero algo trading regulatory burden.
Furthermore, SEBI's 2024 Consultation Paper on "AI/ML in Securities Markets" (released October 2024) explicitly addressed the growing interest in AI-driven trading tools. The consultation identified three key concerns: (1) explainability of AI decisions, (2) protection against hallucination-driven losses, and (3) audit trail requirements. SAHI's deterministic architecture directly addresses all three concerns because the execution layer contains zero AI — every action is the result of a transparent, user-approved IF-THEN rule tree that generates a complete audit trail. This positions SAHI's architecture as potentially more compliant than any future SEBI requirements, rather than scrambling to add compliance retroactively.
Every agent action is visible in the Activity Feed with full audit trail. The agent's decision logic is a simple IF-THEN rule tree that was reviewed by the user before activation. There are no neural network weights to explain — just deterministic conditions. This directly satisfies SEBI's explainability requirement from the 2024 AI consultation paper.
Financial-grade infrastructure with no external API key sharing, no third-party order routing, and fully sandboxed agent execution. All market data processing happens within SAHI's existing compliant infrastructure. User authentication via 2FA + device binding. Compliant with IT Act 2000 and SEBI's Cybersecurity Framework (SEBI/HO/MIRSD/CIR/PB/2018/147).
Mandatory user approval for every agent before activation. Users can edit, pause, or kill any agent at any time. Hardware kill switch to immediately deactivate ALL agents and close all positions within 100ms. SEBI-mandated risk disclosure for all F&O agents. Granular consent management — users explicitly acknowledge each agent's risk profile.
The SEBI Algo Trading Registration Question. A critical regulatory consideration is whether SAHI's agentic system requires registration as an algorithmic trading system under SEBI's framework. The answer depends on the phase of implementation. In Phase 1 (Smart Alerts+), the system only generates enhanced GTT (Good Till Triggered) orders — these are already a standard brokerage feature and do not require algo registration. In Phase 2 (Auto-Pilot), where simple auto-execution begins (SIP automation, cash sweeps, trailing stops), the system may need to be classified as a "semi-automated" system. SEBI's current framework has grey areas here, and proactive engagement with SEBI through the sandbox mechanism (SEBI Innovation Sandbox) is recommended. In Phase 3-4 (Full Agentic / Autonomous), where multi-leg F&O strategies are executed automatically, algo trading registration with NSE/BSE will likely be required.
The phased approach is specifically designed to manage this regulatory escalation. SAHI can launch Phase 1 immediately with zero regulatory risk, begin Phase 2 conversations with SEBI while Phase 1 generates user traction and data, and apply for full algo registration (if required) before Phase 3 launch. This gives SAHI a 12-18 month runway to shape the regulatory conversation rather than reacting to it — a significant strategic advantage.
These six guardrails are hard-coded into the execution engine and cannot be overridden by the AI, the user, or any configuration change. They exist as the final safety net — even if every other system fails, these guardrails will prevent catastrophic losses. They are implemented at the infrastructure level, not the application level, making them immune to software bugs in the agent logic.
User sets maximum daily and weekly loss thresholds. If any agent's cumulative realized + unrealized loss exceeds these thresholds, ALL agents are automatically paused and the user is notified via push + SMS + Telegram. The cap is evaluated on every tick — not on a polling interval.
No single agent can accumulate positions exceeding X% of portfolio value (configurable, default 10%). This prevents concentration risk from any single agent's strategy, regardless of how many triggers fire. Evaluated pre-order — orders that would breach the limit are rejected before reaching the exchange.
Optional "confirm before execute" mode where the agent proposes a trade via push notification and waits for user confirmation (tap-to-approve) before placing the order. Available for all agent types. Defaults to ON for all F&O agents in Phase 2. SEBI-aligned user consent mechanism.
Mandatory acknowledgment screen before activating ANY F&O agent. Shows: historical loss rates for the strategy type, SEBI's standard F&O risk disclosure, and a personalized risk score based on the user's portfolio and leverage. Per SEBI Circular SEBI/HO/MIRSD/MIRSD-PoD-1/P/CIR/2024/37 on investor protection.
Every condition check, trigger evaluation, order placement, fill confirmation, and error event is logged to an append-only, immutable audit trail. Timestamps are sub-millisecond. Data retention: minimum 8 years (per SEBI record-keeping requirements under Securities Contracts Act). Exportable for SEBI inspection.
Panic button accessible from app home screen, notification panel, and dedicated widget — one tap to immediately deactivate ALL agents, cancel ALL pending orders, and optionally close ALL open positions at market. Execution guaranteed within 100ms. Can also be triggered via SMS ("KILL" to registered number) for scenarios where the app is inaccessible.
The implementation roadmap is designed as a 4-phase escalation that maps directly to SEBI's regulatory thresholds. Each phase adds capability while managing compliance risk. Phase 1 launches with zero regulatory burden (enhanced alerting), Phase 2 introduces simple auto-execution that may need algorithmic trading disclosure, Phase 3 delivers the full agentic experience with potential algo registration, and Phase 4 achieves full autonomy with comprehensive SEBI sandbox alignment. This graduated approach means SAHI can start generating revenue and user feedback from Phase 1 within 3 months while the more complex regulatory conversations proceed in parallel.
The bottom line: SAHI already has the hardest parts — the broker license, the OMS, the real-time data infrastructure, and critically, the proprietary derivatives intelligence that creates an unassailable data moat. The agentic layer is the fastest part to build. And with India's ₹200L crore/month F&O market as the trigger engine, it creates a competitive moat that no other Indian brokerage can replicate in less than 18-24 months. The question is not whether agentic brokerage will arrive in India — it is whether SAHI will be the one to define the category.
Natural language → GTT/alert creation. Multi-condition alerts combining price + OI + PCR + VIX. No auto-execution — enhanced alerting only. Users describe what they want to monitor, the AI creates the alert configuration, and SAHI's existing notification system delivers the alerts.
Simple auto-execution for low-risk, well-understood operations: SIP automation, idle cash sweeps to liquid ETFs, trailing stop-loss management, calendar-based rebalancing. Manual approval (confirm-before-execute) mode available and default ON for any F&O-related execution.
Complete NLP → Clarification → Workflow → Approval → Deterministic Execution loop. Multi-leg F&O strategies (straddles, covered calls, iron condors). Human-in-the-loop required for activation. Backtesting with 6-month historical simulation before any agent goes live.
Fully autonomous agents operating 24/7 with comprehensive guardrails. OI-driven agents, covered call income generators, portfolio rebalancers, fundamental screen agents, tax optimizers — all running continuously with real-time condition monitoring.