KX + WorkingAgents: Governed Agentic AI for Capital Markets

KX builds the world’s fastest time-series database and vector analytics platform. WorkingAgents governs what AI agents are allowed to do with it. When autonomous agents are making trading decisions, analyzing market signals, and executing research workflows at sub-second latency, ungoverned access is not a risk you can absorb. The partnership: KX provides the high-performance data engine, WorkingAgents provides the control plane.

What KX Does

KX is the company behind kdb+, independently benchmarked as the fastest in-memory columnar analytics database in the world. Founded in 1993, they have spent three decades building the infrastructure that powers the most data-intensive operations in finance, defense, and manufacturing.

Core Products

kdb+ – The original. A high-performance time-series database and analytics engine powered by the vector language q. Ultra-low latency access to real-time and historical data. Used by the world’s top investment banks, hedge funds, and trading firms for tick data capture, high-frequency trading, backtesting, risk analytics, and market surveillance.

KDB-X – The next generation. Unifies time-series, vector, and AI workloads in one platform. Built-in AI libraries, native interoperability across Python, SQL, and q. Designed for teams building and deploying modern analytics and AI workflows at scale. General commercial availability in early 2026.

KDB.AI – A multi-modal vector database enabling scalable, real-time AI applications with advanced search, personalization, and Retrieval Augmented Generation (RAG). Integrates with NVIDIA cuVS for GPU-accelerated vector indexing and search.

The NVIDIA Partnership

KX is deeply embedded in the NVIDIA AI stack. At GTC 2026 (March 16-19, San Jose), KX is launching two agentic AI blueprints built on NVIDIA accelerated computing and NVIDIA AI Enterprise software:

AI Research Assistant – Accelerates research workflows by improving retrieval, summarization, and synthesis across structured market data, unstructured content, and proprietary documents. Designed for earnings analysis, regulatory filings, and investment research.

Trading Signal Agents – Helps teams discover, validate, and monitor trading signals in real time. Aligns multi-modal information in market time, produces governed outputs for trading and risk processes.

The technical stack includes NVIDIA NeMo Retriever, Nemotron embedding models, NIM microservices, and KDB.AI’s integration with NVIDIA cuVS. The key differentiator is temporal AI – ensuring AI systems understand not just what happened, but when it happened, computing point-in-time correct context for auditable, repeatable workflows.

Industries and Customers

KX is trusted by the world’s leading investment banks and hedge funds. Beyond capital markets, they serve aerospace and defense, life and health sciences, semiconductor manufacturing, telecommunications, energy, and automotive. Notable partnerships include Intercontinental Exchange (ICE) for real-time analytics across 25+ million financial instruments, and Databricks for integrated time-series analytics.

Synergies with WorkingAgents

KX is building agentic AI systems that make autonomous decisions in capital markets – one of the most regulated, highest-stakes environments in the world. These agents need governance. WorkingAgents is purpose-built for exactly this problem.

1. Permission-Scoped Trading and Research Agents

KX’s agentic blueprints give AI agents the ability to search market data, analyze signals, execute research workflows, and produce trading recommendations. Different roles need different access:

Senior Quant Analyst Agent
  Read: full tick database, proprietary research, all signal libraries
  Write: trading signal validation results, research summaries
  Execute: backtesting workflows, cross-asset correlation analysis

Junior Research Agent
  Read: public filings, earnings transcripts, curated market data
  Write: research drafts (flagged for review)
  Execute: single-asset analysis only
  Blocked: proprietary signal libraries, trading execution

Compliance Agent
  Read: all audit trails, all agent actions, all data access logs
  Write: compliance flags, investigation reports
  Execute: surveillance queries
  Blocked: trading execution, signal modification

WorkingAgents’ Virtual MCP Servers enforce these boundaries through configuration, not code. Each agent inherits its user’s permissions – one identity, one set of rules, full accountability.

2. Three-Checkpoint Guardrails for Market-Speed AI

KX’s systems operate at sub-second latency. WorkingAgents’ guardrails operate at all three checkpoints without adding meaningful overhead:

Pre-execution – Validate that an agent’s query parameters are within its permitted scope before it touches the kdb+ database. Block unauthorized cross-asset queries, prevent unauthorized access to proprietary signal libraries, catch prompt injection attempts before they reach data systems.

Real-time – Human-in-the-loop for high-risk operations. “The trading signal agent wants to flag AAPL for position adjustment based on earnings sentiment – approve or deny?” Low-risk research queries execute autonomously within defined guardrails.

Post-execution – Inspect outputs before they reach downstream systems. Redact PII from research summaries, mask proprietary signal parameters in shared reports, filter confidential trading strategies from cross-team communications.

3. Audit Trails for Regulated Markets

Capital markets regulators (SEC, FCA, MAS, ESMA) require knowing exactly what happened, when, and why. KX’s temporal AI ensures point-in-time correctness. WorkingAgents provides the unified audit trail across every system the agent touches:

{
  "workflow": "trading-signal-analysis",
  "agent": "quant-research-senior",
  "user": "[email protected]",
  "steps": [
    {
      "tool": "kx.kdbai.vector_search",
      "query": "earnings sentiment AAPL Q1 2026",
      "guardrails": {"scope_check": "passed", "data_classification": "proprietary"}
    },
    {
      "tool": "kx.kdbx.signal_validation",
      "result": "signal confirmed, confidence 0.87",
      "guardrails": {"human_approval": "not_required", "risk_threshold": "within_bounds"}
    },
    {
      "tool": "kx.research.generate_summary",
      "result": "research note generated",
      "guardrails": {"pii_check": "passed", "proprietary_filter": "applied"}
    }
  ],
  "total_latency_ms": 340,
  "audit_hash": "sha256:a4f2..."
}

Every action is logged, timestamped, and cryptographically verifiable. When the regulator asks what the AI agent did and why, you have the answer.

4. Compliance-Ready Architecture

KX serves the most regulated industries on earth. WorkingAgents is designed for SOC 2, HIPAA, GDPR, and FedRAMP compliance:

5. MCP Gateway as the Bridge

WorkingAgents’ MCP Gateway with 86+ governed tools becomes the integration point. KX’s kdb+ and KDB.AI services register as tools in the gateway. Every agent that accesses KX infrastructure goes through the gateway’s permission and guardrail layer. Centralized token management replaces scattered credentials across trading desks and research teams.

Value Proposition

What KX Gains

What WorkingAgents Gains

GTC 2026 Approach

KX is at Booth #186 at NVIDIA GTC 2026 (March 16-19, San Jose). They are presenting “From Signal to Strategy: Unlock Alpha With AI Powered Research and Trading” on Monday, March 16, at 5:00 PM PT.

Conversation Starter

“Your agentic blueprints solve the intelligence problem – agents that understand temporal market data and produce real trading signals. We solve the governance problem – making sure those agents only access what they’re authorized to, every action is auditable, and the whole thing is compliant enough for a Tier 1 bank’s risk committee. We’d like to explore what it looks like to put a governance layer around your blueprints so they’re deployable in regulated environments on day one.”

Key Talking Points

  1. The governance gap in agentic AI for capital markets. KX has built remarkable agent capabilities. Banks will ask: who controls what these agents can access? What’s the audit trail? How do we prove to the FCA that the AI agent was operating within authorized parameters? WorkingAgents answers all three.

  2. Virtual MCP Servers map to trading desk structure. Each desk – equities, fixed income, FX, commodities – gets its own permission boundary. The equities research agent cannot see the FX desk’s proprietary signals. Configuration, not code.

  3. One instance per customer eliminates data commingling risk. For hedge funds and prop trading firms, this is non-negotiable. Their alpha is in their data. WorkingAgents’ deployment model guarantees complete isolation.

  4. Three-checkpoint guardrails at market speed. Pre-execution validation, real-time human-in-the-loop for high-risk decisions, post-execution output filtering – all without adding meaningful latency to KX’s sub-second workflows.

What to Ask Them

What to Offer

WorkingAgents is an AI governance platform specializing in agent access control, orchestration, and security for enterprises deploying AI at scale.