November 2026 Summaries

Kong plans to participate in AWS re:Invent 2026 in Las Vegas from November 30 to December 4, presenting its platform for managing API, AI, and event traffic, with an emphasis on connectivity, governance, observability, metering, and security for AI systems. At booth #1485 in The Venetian, attendees can see demonstrations of Kong’s AI Gateway, Event Gateway, Konnect Metering and Billing, and Konnect Catalog, which support oversight of models, agents, MCP servers, Kafka-based data traffic, token spending, and approved AI resources. The company will also host a November 30 session with FanDuel leaders on API foundations for innovation, offer meetings with its executives, and participate in several invitation-based networking and lunch events focused on AI governance, security, shadow AI, and emerging threats alongside organizations including AHEAD, Harness, Infosys, and Noma Security.

llm-d is an open-source, Kubernetes-native routing framework that improves the efficiency of self-hosted LLM inference engines such as vLLM without running models itself. The post contrasts self-hosting with managed APIs, noting that self-hosting provides control, data locality, predictable performance, and customization opportunities but requires teams to operate GPUs, scheduling, authentication, rate limiting, and observability. It demonstrates an Amazon EKS deployment in which vLLM serves Llama 3.2 3B on NVIDIA GPUs, llm-d routes requests among model-server pods using workload-aware policies, and Kong AI Gateway provides the public endpoint, API-key authentication, model aliases, token-based rate limits, usage reporting, and Konnect observability. llm-d’s key capabilities include routing requests to replicas with matching KV-cache prefixes, separating compute-heavy prompt prefill from memory-bound token decoding, and favoring replicas that already host requested LoRA adapters. The first part establishes the infrastructure and basic request flow, while a planned second part expands the deployment to multiple replicas and evaluates cache-aware routing, prefill/decode disaggregation, and adapter serving.

Kong Event Gateway 1.3 adds policy-based controls for shared Kafka environments while operating in front of existing clusters through Kong Konnect. Its Request Rule Validator inspects Kafka request contents, allowing teams to enforce requirements for topic partitions, replication factors, naming, ACL bindings, and producer settings, with violations either rejected using Kafka’s POLICY_VIOLATION error or logged while requests proceed. New record transcoding policies convert JSON to Avro for producers and Avro to JSON for consumers, reducing the need for client-side schema registry integration or custom connectors while retaining Avro in Kafka. The Mask Fields policy redacts selected string fields during production or consumption, supports several masking methods, and can use schema-registry metadata such as PII tags to automatically protect newly tagged fields, though masking may affect client validation and does not yet support numeric or Boolean fields. The release also enables gateway-to-Amazon MSK authentication through AWS IAM credentials and role assumption, separately from client authentication. Kong positions these features as a way to provide self-service Kafka governance, data protection, and format compatibility as more applications and AI agents access event streams.

Kong AI Gateway and a FinOps platform can be combined to provide real-time AI spending visibility and monthly invoice-based chargeback without requiring user or cost-center tagging in cloud providers. The gateway meters model requests across providers such as OpenAI, Anthropic, AWS Bedrock, Google Vertex AI, and Azure OpenAI, attributes usage to users, agents, applications, and cost centers, applies negotiated token rates, and can enforce budget alerts or hard token and dollar limits. After month-end, a true-up process groups gateway records and invoice data by month, provider, model, and region, calculates settlement factors from actual billed amounts, and adjusts estimated per-user costs so chargebacks match invoices while unmatched fixed costs, credits, or bypassed traffic remain in a central AI category. Implementation involves routing AI traffic through the gateway, exporting Kong usage data, connecting cloud and vendor billing feeds, mapping provider SKUs to models, and validating the reconciliation through a pilot team before using settled costs for chargeback.

Kong AI Cost Management is a new capability within Kong AI Management designed to help enterprises track, govern, and optimize spending across AI providers, models, applications, teams, agents, and customers. It uses consumption data and contextual information captured through Kong AI Gateway to provide granular cost attribution beyond the account- or model-level reporting typically offered by providers, enabling organizations to allocate costs, establish accountability, and set budgets with threshold alerts. The platform also includes infrastructure-level optimization tools such as intelligent model routing, semantic and prompt caching, prompt compression, and reasoning controls to reduce token use and redundant model calls. An upcoming Opportunities feature will help teams identify efficiency improvements and measure the savings produced by optimization efforts. By centralizing visibility, budgeting, runtime optimization, and savings measurement, Kong aims to give platform, engineering, finance, and FinOps teams a more scalable way to apply financial discipline to expanding AI and agentic workloads.

Kong has launched Token Vault in Kong AI Gateway 2.3 to let enterprise AI agents securely access OAuth-protected MCP servers and APIs such as GitHub, Slack, Salesforce, Google, and Datadog on behalf of individual users. Users authenticate through their existing corporate OIDC identity provider, while the gateway enforces authorization, logging, rate limits, and other policies before obtaining and injecting a per-user upstream OAuth token, which agents never see. On initial access, users complete an upstream OAuth consent flow through an enrollment link or a self-service Credential Manager, after which the gateway stores encrypted refresh tokens and automatically retrieves, refreshes, and injects credentials for approved requests. Token encryption keys remain available only to customer-controlled gateway data planes, while Kong Konnect can manage encrypted token records without being able to decrypt them. Designed for high-volume, per-user token lifecycle management rather than general secrets storage, the service supports shared credentials across agents and MCP deployments, immediate policy-based access removal, and OAuth client configurations including confidential clients, PKCE, and dynamic client registration.

Kong AI Registry is a centralized platform for registering, discovering, and governing enterprise AI models, Model Context Protocol servers, and AI agents as organizations expand from isolated AI applications to interconnected agentic systems. It aims to replace brittle, manually configured integrations with dynamic discovery of approved capabilities, allowing developers and agents to find reusable resources without requiring every connection to be hard-coded. The registry distinguishes discoverability from authorization, integrating with Kong’s identity, access management, and runtime security tools to ensure that agents can only invoke resources they are permitted to use. Kong AI Gateway provides runtime authentication, authorization, policy enforcement, traffic controls, and observability, while Kong Token Vault lets agents access downstream services without holding underlying credentials. By making approved enterprise resources easier to locate and reuse, the platform is intended to reduce fragmented integrations and shadow AI while maintaining security, governance, and operational visibility at scale.

Kong has introduced AI Management in Kong Konnect to address operational challenges as AI agents move into production, including untracked spending, insecure credential handling, duplicated API-to-tool integrations, and limited visibility into agent activity. AI Cost Management centralizes provider pricing, supports custom rates and spend attribution by team, application, or harness, integrates with third-party AI tools, and enables budget controls and dashboards. Token Vault keeps credentials out of agent code and prompts while brokering access to MCP providers, enforcing least-privilege policies, maintaining audit trails, and allowing targeted revocation. Context Mesh converts existing APIs and MCP servers into governed, reusable MCP tools, supports composite services and a shared catalog, and uses code mode to reduce context and token consumption. AI Observability provides dashboards, querying, request logs, and planned session-level tracing for examining agent steps, tool calls, and model calls. These capabilities are included with Kong AI Gateway on Konnect without a separate SKU.

AI observability extends traditional request-level monitoring by tracing complete AI tasks across model calls, agents, tools, MCP interactions, policies, retries, and handoffs, allowing teams to assess task quality, cost, latency, and failures that may not appear in individual successful requests. Kong AI Observability, integrated into the connectivity path, aggregates signals including token use, spending, latency, errors, retries, cache status, and inference inputs and outputs across sessions, which represent the full journey behind an outcome. Its dashboards, Explorer tool, filtering, exports, and role-based access controls help organizations analyze usage and performance by model, tool, team, application, environment, or business unit, while upcoming session-level tracing will enable turn-by-turn inspection of conversations to identify where issues arose. The platform also supports OpenTelemetry and metric exports for integration with existing observability systems and connects session data with AI cost-management, showback, and chargeback capabilities.

Noah Koeck’s profile traces his path from growing up in rural Austria and helping local businesses with social media marketing to becoming a Commercial Account Executive at Kong serving companies across the DACH region. After joining Kong in London as a Sales Development Representative, he pursued opportunities beyond his formal role, gaining experience in later-stage sales cycles and earning recognition including Global SDR of the Year, Sales Representative of the Year, and a President’s Club qualification alongside the account executives he supported. He attributes his advancement to initiative, continuous learning, feedback, and collaboration across teams, emphasizing that successful sales requires coordinated expertise rather than individual effort. In his current role, Koeck helps organizations use Kong’s API management and connectivity platform to connect applications, AI models, and autonomous agents securely, as AI-related concerns such as cost, risk, and speed to market become executive-level business priorities.