Benchmark

Designed for real-time AI agent retrieval, Lightning+ was evaluated under a demanding workload of 1,000,008 structured data queries, each consisting of a single, complex SQL query. The query randomized the target SYMBOL and selected a dynamic date range of up to 1 year. Despite this variability, Lightning+ demonstrated the ability to deliver low-latency, high-throughput responses. The entire benchmark was completed in under 20 minutes.

The benchmark queries were executed against a preloaded in-memory dataframe containing S&P 500 daily stock data spanning from January 2000 onward.

Rows: 2,234,876 Format: Arrow
Tables Engine: Embedded DuckDB

SELECT
  SYMBOL,
  DATE,
  CLOSE,
  100 * (CLOSE - FIRST_VALUE(CLOSE) OVER w) / FIRST_VALUE(CLOSE) OVER w AS RETURN_PCT
FROM sp500_daily
WHERE 
  SYMBOL = ?
  AND DATE BETWEEN ? AND ?WINDOW w AS ( );


FROM sp500_daily
WHERE
  SYMBOL = ?
  AND DATE BETWEEN ? AND ?
  WINDOW w AS (
  PARTITION BY SYMBOL
  ORDER BY DATE
  ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW
);

This query randomly selects a SYMBOL and generates a dynamic 1-year date range to calculate return percent.

This histogram visualizes the latency distribution of 1,000,008 AI agentic queries processed by Lightning+. Over 95% of responses complete in under 40 milliseconds, with a sharp median at 25 ms.

Key Latency Metrics:

• Median (P50): 25 ms
• P75: 31 ms
• P90: 36 ms
• P95: 40 ms
• Maximum Latency Observed: 80 ms

Benchmark Duration: ~19 minutes
Sustained Throughput: ~52,600 responses per minute

These results validate Lightning+ as a consistent, real-time retrieval engine for structured data.

Modern AI agents — from copilots to planners — depend on fast, accurate access to structured data. Yet most data infrastructure was built for dashboards, not decisions. When retrieval is slow or stale, agents:

• Hallucinate
• Drop context
• Fail to act in real time

Modern AI agents — from copilots to planners — depend on fast, accurate access to structured data. Yet most data infrastructure was built for dashboards, not decisions. When retrieval is slow or stale, agents:

• Hallucinate
• Drop context
• Fail to act in real time

Lightning+ changes the game.

Built for real-time AI workloads, it eliminates ingestion delays and cloud latency by keeping data local, columnar, and query-ready.

Key Takeaways from the Benchmark

• 1,000,008 complex queries completed in under 20 minutes
• Median response time: 25 ms
• 95% of queries completed in < 40 ms
• Sustained throughput: 52,600+ queries per minute
• Max latency: under 100 ms — even under load

Why It Matters for Production AI

• Instant access to live, structured data — no prep pipelines required
• Consistent performance keeps agents responsive and grounded
• Edge-optimized — no egress, no cloud dependency, full control

Lightning+ is not just fast — it’s ready for agent-scale AI