Broker API reliability for live algo trading India

Why broker API reliability matters for live algos

Live algorithmic trading in India depends on a broker API for every critical action - placing orders, modifying or cancelling them, and confirming fills. Social discussions repeatedly highlight a simple point: a strategy can be profitable on paper, but fail in production if the broker platform is down. Reliability issues show up as dropped WebSocket connections, delayed order updates, and intermittent errors during volatile periods. Traders also flag rate limits as a reliability issue, because throttling can block a strategy at the exact moment it needs to react. Latency is discussed differently depending on the strategy type, with many retail traders saying 100-500 milliseconds is adequate for swing and positional approaches. The same threads note that scalping and complex multi-leg execution can become sensitive to sub-100ms delays. Several posts also stress that a broker choice is not just about low fees, but about consistent execution quality and predictable API behaviour. The shared view is that API reliability is an input to risk management, not a convenience feature.

What a “good” broker API looks like in practice

Across Reddit and developer groups, the checklist starts with REST endpoints for order execution and WebSocket streaming for real-time market data. Documentation quality is treated as a core feature because poor docs translate into production bugs and longer debugging cycles. Rate limits matter once you run more than one strategy, or when your logic needs to fire multiple orders quickly. WebSocket support is widely considered mandatory for live price feeds, but traders also warn you must build reconnection logic because disconnects happen. Historical data depth comes up because SEBI-aligned workflows emphasise backtesting and risk validation before going live. A well-maintained Python library is frequently cited as a practical advantage, because it reduces the chances of implementation errors in authentication and order workflows. Authentication details are also treated as reliability-critical, especially where session tokens expire daily and must be refreshed. Finally, cost is framed as part of reliability budgeting, because some APIs charge a monthly fee regardless of usage.

Quick comparison of broker APIs discussed by traders

The table below consolidates points repeatedly referenced in social threads, including costs and limits where users provided them. Where users disagree on pricing, the table reflects that uncertainty instead of forcing a single number. Execution and stability comments are based on user-reported experience shared in the same discussions. This is not a performance benchmark, but a summary of what retail algo traders are using and debating. The key takeaway from the comparisons is that “free” does not automatically mean “good enough” for every style of automation. Similarly, a paid API is not automatically better if it does not match your workflow, limits, and compliance requirements. Traders also stress that the developer ecosystem around a broker can be as important as raw features. Use the table as a starting point, then validate with your own latency and error-rate tests.

Zerodha Kite Connect: mature ecosystem with strict limits

Zerodha is repeatedly described as the dominant retail API in India, largely because Kite Connect is mature and widely adopted. Traders cite clear documentation, an active developer community, and a well-maintained Python library as major practical benefits. The API supports order placement, WebSocket-based live market data, historical OHLC data, and portfolio and account endpoints. Multiple posts highlight the 3 orders/second rate limit and frame it as enough for most retail strategies, but restrictive for complex multi-leg execution. Users also point out the daily token expiry pattern, which makes automated token refresh non-negotiable for production systems. WebSocket streaming is described as reliable and low-latency, but still requires connection management and reconnection handling. The main drawback discussed is cost, at around ₹2,000 per month, which can matter for early-stage traders. Despite the fee, many users argue its reliability and ecosystem justify it for serious live deployment.

Angel One SmartAPI and Fyers: free access with trade-offs

Angel One’s SmartAPI is often presented as the strongest free API option in these discussions. Users cite 10 orders/second rate limits and reliable WebSocket streaming as helpful for multi-strategy portfolios. Authentication is described as token-based and simpler than OAuth flows, which reduces integration friction for new developers. At the same time, some traders report occasional reliability issues during high market volatility, which can be a factor for fast intraday automation. Fyers is discussed less as an execution-first choice and more as a backtesting-friendly API option. The standout claim is free minute-level historical OHLCV data going back 1-2 years, which users say can reduce dependence on paid third-party data. Threads also mention Fyers has a community focus and trader-centric features that appeal to developers. The common theme for both is that “free” is a strong advantage, but production readiness still depends on your monitoring, retries, and risk controls.

Dhan, Upstox, Kotak, Groww, ICICI: where traders place them

Dhan is frequently mentioned for option analytics and a modern UI, with users calling out webhook automation as part of the workflow. Pricing is discussed around ₹500 per month, sometimes framed as a data pack, and it is grouped with “cheap APIs” rather than fully free. Upstox is described as having a detailed developer API with real-time execution and live data streaming, but some users note occasional lag. Upstox authentication is also discussed as having longer-lived tokens compared with brokers where tokens expire daily, which can reduce mid-session failures if implemented correctly. Kotak Neo is positioned as a fast-execution option, with claims of sub-50ms latency and high order throughput backed by full-service infrastructure. Some posts also highlight zero brokerage on API orders under specific “trade free” plans, which makes it attractive for cost-sensitive automation. Groww is described as entering algo trading with performance-first APIs, citing around 65ms average response times and explicit request-per-minute limits for orders and market data. ICICI Direct’s Breeze API is discussed as a brand-trust choice, with useful data endpoints including options chain information, but an experience that can feel dated and potentially slower than discount brokers.

SEBI framework in 2026: Algo-ID, limits, and auditability

A major theme in these discussions is that reliability now includes compliance readiness, not just uptime. Users refer to SEBI’s 2024-25 framework requiring broker-level validation, risk controls, and audit trails for algo-generated orders. Posts emphasise mandatory tagging of algo orders with a unique algo identifier, and that unapproved API-based connectivity can lead to restrictions. Several threads state that from April 2026, every algorithm must be registered via the broker for an Algo-ID before live execution. A separate point discussed is an order-per-second threshold around 10 OPS, where exceeding it triggers additional compliance scrutiny. Traders also note that even simple “systematic” scripts may fall under the same Algo-ID tagging expectation if they place orders automatically. Record-keeping is repeatedly highlighted, with a stated need to maintain detailed order and execution logs for at least five years. The practical implication is that broker selection should include how streamlined the approval process is and how clearly the broker supports API-based algo trading within SEBI’s framework.

Production reliability checklist traders keep repeating

Operational reliability shows up in small engineering details that can stop trading silently if ignored. Token handling is the most common example, especially for APIs where tokens expire daily and require refresh before market hours. Traders recommend secure storage of keys and tokens, avoiding hardcoding, and using environment variables or encrypted configs. WebSocket disconnects are treated as normal, so reconnection logic with exponential backoff is widely recommended. Several posts suggest queueing orders during disconnections, then reconciling once connectivity returns, to avoid partial state updates. Error handling practices include catching exceptions, logging them with timestamps, and retrying a limited number of times rather than looping indefinitely. Risk controls are repeatedly described as mandatory in practice, including position limits, maximum order size, and a daily loss kill switch that stops new orders. Another recurring suggestion is order de-duplication, where the system checks recent orders before sending a new one to avoid repeats during retries. Traders also emphasise alerting, because an API can fail in a way that does not crash your script but still blocks orders.

Choosing by strategy frequency, not just by brokerage

Social discussions draw a consistent line between retail automation and high-frequency trading, saying retail APIs are not meant for co-located HFT. Many posters argue that if your strategy needs more than a few orders per second consistently, your design is likely outside the intended retail API use case. For swing and positional strategies, the same users say latency in the 100-500ms range is usually acceptable, shifting attention to uptime, order-state accuracy, and clean reconciliation. For intraday automation, traders focus more on rate limits, WebSocket stability, and how the API behaves during volatile periods. Cost comparisons often miss the bigger issue: hidden “costs” from missed trades due to throttling, lag, or unhandled token expiry. Developer ecosystem also matters because a large community means more examples, faster debugging, and fewer integration surprises. SEBI compliance adds another filter, because approval flows, Algo-ID tagging, and audit logs affect how quickly you can go live and how safely you can operate. The common advice is to shortlist two brokers, run paper trading, and validate reliability under load before risking capital. For 2026, threads increasingly treat broker selection as an engineering decision plus a compliance decision, not a pricing decision.