QueryPilot — Catch Database Regressions Before Production

Deterministic Performance

More Scale, Less Latency

QueryPilot inspects execution plans with zero database locks. Detect slow scans, generate optimal covering indexes, and fail pull requests before production regressions occur.

CRITICAL: Seq Scan Detected

└─ Parallel Seq Scan on orders

Filter: (created_at > '2026-01-01')

cost: 0.00..41890.50 | rows: 184,000

└─ Hash Join (users.id = orders.user_id)

Identifies full table scans over 1.4M rows. Computes startup vs total cost ratio deterministically.

AI Index Strategy Ready

-- Recommended DDL Migration

CREATE INDEX CONCURRENTLY

idx_orders_status_covering

ON orders (status, created_at DESC)

INCLUDE (user_id, amount);

⚡ Estimated Speedup: +98.6% Faster (Cost dropped from 41,890 → 12.4).

CI GATE FAILED (Threshold 20%)

Query	Base	PR	Diff
get_active	12.4	12.8	+3.2%
list_trans	45.1	8,420	+18,569%
user_feed	88.0	89.1	+1.2%

Halts pipeline with exit code 1 when execution plan cost exceeds the baseline threshold.

Security & Policy Control

ArmorIQ Zero-Trust Governance

When autonomous AI agents interact with your database, ArmorIQ validates cryptographic CSRG intent tokens to approve read-only inspections while blocking destructive schema mutations.

Cryptographic Intent Verification

Autonomous agents can easily generate dangerous DDL or unrestricted table locks. QueryPilot embeds ArmorIQ at the control plane to enforce strict zero-trust boundaries.

CSRG Intent Tokens

Mints short-lived (300s TTL) cryptographic tokens required for all tool invocations.

Read-Only Tool Authorization

Enforces policy allowing explain_query and analyze_plan while denying raw writes.

Audit Logging & Replay Trail

Every decision is signed and verifiable via the ArmorIQ customer proxy.

Config: armoriq.yaml · Policy Engine Active

armoriq.yaml

version: v1
identity:
  api_key: $ARMORIQ_API_KEY
  user_id: querypilot-user
  agent_id: querypilot

environment: sandbox
proxy:
  url: https://customer-proxy.armoriq.ai
  timeout: 30

mcp_servers:
  - id: querypilot-db
    auth:
      type: bearer
      token: $QUERY_PILOT_MCP_TOKEN

# Zero-trust principle: allow inspection, block mutation
policy:
  allow:
    - querypilot-db.explain_query
  deny: []

intent:
  ttl_seconds: 300
  require_csrg: true

Agentic Tooling

Model Context Protocol Server

QueryPilot operates an MCP server over Streamable HTTP on port :3333. Seamlessly connect Claude Desktop, Cursor, or autonomous agents to analyze database plans.

Tool: explain_query

Safe Plan Extraction

Accepts raw SQL queries, runs EXPLAIN (FORMAT JSON) without executing data mutations, and returns parsed AST metrics.

Tool: analyze_plan

Anti-Pattern Detection

Scans the execution tree for high-cost sequential scans, cartesian products, and missing foreign key join indexes.

Tool: suggest_indexes

Deterministic DDL

Cross-references table statistics and schema to output exact CREATE INDEX CONCURRENTLY recommendations.

claude_desktop_config.json

{
  "mcpServers": {
    "querypilot": {
      "command": "npx",
      "args": ["tsx", "src/mcp/server.ts"],
      "env": {
        "PORT": "3333",
        "DATABASE_URL": "postgresql://..."
      }
    }
  }
}

Hands-on Sandbox

Live Execution Lab

Test real SQL queries against our plan evaluator. Switch presets or write custom SQL to see live cost calculation and AI index generation.

SQL Query Input

CRITICAL: Seq Scan Cost: 41,890.50

Execution Plan Tree:

└─ Parallel Seq Scan on orders

Filter: (status = 'unfulfilled' AND created_at >= '2026-01-01')