September 2026 Summaries

Kong announced an evolution of Kong Konnect into an AI Connectivity Platform intended to help enterprises deploy and govern production AI agents across their full lifecycle. The platform addresses six operational needs: reacting to business events, discovering approved tools and services, securely accessing credentials, efficiently consuming context, tracking costs, and providing end-to-end observability. New and expanded offerings include Webhook Engine for event-triggered agent workflows, an Agent & MCP Registry and the generally available Konnect Catalog for governed discovery, Token Vault for credential-free agent access, and the generally available Context Mesh for converting APIs into agent-ready MCP servers. Kong also introduced early-access AI Cost Management for attributing model and agent spending and Advanced AI Observability for tracing and debugging entire agent conversations. Konnect is designed to apply consistent governance across cloud, on-premises, edge, and self-hosted environments, emphasizing portability, data residency, and reduced vendor lock-in.

Kong is expanding its AI Gateway cost-management capabilities to help enterprises measure, attribute, govern, and optimize AI spending as usage spreads across models, providers, applications, teams, and autonomous agents. The company argues that token- and provider-level reporting alone does not explain the business purpose of AI costs, so its gateway calculates request-level pricing and connects consumption to organizational context such as workflows, users, products, teams, regions, and cost centers. This attribution enables organizations to relate AI spending to operational outcomes, establish budgets aligned with business units, identify unusual or growing usage in real time, and enforce consumption policies before costs exceed expectations. Kong also offers optimization methods including semantic caching, prompt compression, and intelligent model routing, while measuring realized savings and identifying further cost-reduction opportunities. By separating business attribution from underlying AI infrastructure, the platform aims to preserve consistent financial governance even as organizations change models, providers, pricing arrangements, or deployment environments, positioning the gateway as a unified control point for AI security, reliability, and economics.

Kong Webhook Engine, introduced in private beta, is designed to convert business events from sources such as Kafka into HTTP-triggered actions for applications, workflows, SaaS services, and AI agents. It allows systems to register for specific events and receive them through webhook endpoints, reducing the need to build and operate custom Kafka consumers or continuously poll for changes. The platform supports durable buffering and fan-out, enabling one event, such as a suspicious transaction or shipped order, to notify multiple independent destinations including fraud agents, customer systems, analytics tools, and compliance platforms. Kong positions the capability as a bridge between event-driven architecture and agentic AI, allowing agents to act on real-time business context by investigating events, using connected tools, and escalating or automating responses. When used with Kong Event Gateway, Webhook Engine can also apply governance controls such as access management and isolation before approved events are delivered downstream.

Kong has announced the general availability of Context Management within Kong AI Management to help enterprises give AI agents efficient access to APIs, MCP servers, and data sources without overloading model context windows. The offering addresses the drawbacks of exposing every operation in large enterprise APIs as separate MCP tools, which can increase token costs, latency, and tool-selection difficulty while reducing reasoning quality. Its Code Mode uses a compact “search, understand the schema, execute” interface, allowing agents to discover relevant API operations and retrieve schemas only when needed rather than processing hundreds of tool definitions at once. Code Mode also lets agents compose multi-step API workflows with programming logic, potentially reducing repeated model-to-tool interactions. Building on Kong Context Mesh and AI Gateway, Context Management aims to centralize how organizations curate, control, and expose enterprise capabilities so agents can access relevant information and actions efficiently as MCP adoption expands.

Kong AI Registry is a proposed AI-native, governed discovery layer that enables agents to find and use approved enterprise capabilities, including models, MCP servers, agents, tools, resources, and skills, without developers manually configuring every connection. Organizations can register assets with metadata such as ownership, versions, risk, authentication, policies, and access controls, then approve and publish them to audience-specific registries for groups such as engineering, finance, contractors, or partners. The approach separates responsibilities by allowing organizations to define permitted capabilities at build time while agents dynamically discover and consume appropriate resources at runtime. Unlike the human-facing Kong Catalog, which provides a platform-wide inventory of AI, API, and event assets, AI Registry focuses on what particular agents are authorized to discover and use. Kong positions the registry as broader than an MCP server directory and plans to make it generally available at Kong AI Summit 2026.

Kong AI Gateway 2.2 is presented as an update designed to help enterprises govern a rapidly diversifying AI ecosystem without tying their infrastructure to a single model, provider, or interface. It adds native support for TypeSafe JEV, which returns structured probabilistic decisions that can support uses such as model-routing choices, and introduces Skills API support that translates between a common Kong representation and provider-specific implementations. A new passthrough mode allows custom, self-hosted, preview, and non-LLM AI endpoints, including vLLM, Ollama, and NVIDIA NIM, to operate behind the gateway without request-body transformation while retaining authentication, rate limiting, and logging. The release also supports custom plugins for organization-specific policies, expands advanced rate limiting with credential-based matching, and offers a technical-preview integration with Headroom for prompt compression to reduce token use, cost, and potentially latency while tracking compression metrics and preserving fallback behavior. Overall, the update emphasizes “strategic portability,” aiming to give developers flexibility in choosing AI technologies while providing platform teams a consistent layer for security, governance, consumption controls, extensibility, and operational visibility.

Kong has launched the generally available evolution of its Service Catalog into Kong Catalog, expanding its existing API discovery, documentation, versioning, and lifecycle governance capabilities to cover AI models, MCP servers, agents, event streams, and APIs across clouds and gateway environments. The platform is positioned as a centralized system of record intended to address growing AI infrastructure sprawl, allowing teams to identify available interfaces, their owners, and their suitability for reuse or development. AI models and MCP servers are available as first-class catalog entities, agents are currently in beta with general availability planned, and documentation and versioning capabilities for AI entities are expected to follow. Kong also highlights KAi-powered semantic search, which enables users to find resources through natural-language queries. The company contrasts its cloud- and gateway-agnostic approach with provider-specific AI registries and notes that the prior Service Catalog will remain supported without an end-of-life date, although it will receive no new functionality and some features are being redesigned for the expanded API and AI scope.

Kong Identity has introduced Token Vault, a capability designed to let AI agents securely access enterprise systems without directly storing the downstream credentials required by services such as GitHub, Jira, Slack, Salesforce, Snowflake, Databricks, internal APIs, MCP servers, and event platforms. Rather than distributing credentials to agents, organizations can authenticate agents through Kong, apply centralized policies governing permitted resources, and have Token Vault supply the appropriate per-user, shared, OAuth, API-key, or other credential during authorized requests. This approach separates an agent’s identity from the credentials used to access external systems, reducing the risks of credential exposure if an agent is compromised or behaves unexpectedly. It also allows security teams to revoke or modify access centrally, maintain audit records of resource use, and manage heterogeneous authentication models through a single enforcement layer. Token Vault expands Kong Identity’s existing principals directory and access-management features, positioning centralized identity, policy enforcement, and credential brokering as an alternative to embedding more secrets into increasingly capable AI agents.

Kong is previewing advanced AI observability capabilities for Kong Konnect that treat an entire AI session, rather than an individual request, as the primary unit of analysis. The approach is intended for complex, multi-turn agentic workflows involving multiple model calls, tools, MCP servers, policies, retries, and agents, where request-level telemetry may not reveal why a system’s behavior deteriorates. Session-level tracing connects these operations so developers can inspect inputs, outputs, token usage, cost, latency, failures, guardrail execution, and model or tool activity across a complete interaction. Kong also aims to provide unified analysis of performance metrics such as model success rates, time to first token, cache hit rates, outliers, and cost per request or session across AI infrastructure. A forthcoming Konnect Debugger is planned to supplement post-hoc tracing with live investigation capabilities, supporting a workflow of observing, tracing, investigating, and debugging AI behavior in context.

Kong Operator 2.3 expands Kubernetes-native management for Kong’s API, AI, Event Gateway, and Dev Portal offerings, with a particular focus on deploying and operating AI Gateway through GitOps workflows. New custom resource definitions allow teams to declaratively configure AI Gateway control planes, LLM providers and models, routing strategies, authentication, consumers, usage attribution, and security policies such as prompt-injection detection and topic allowlisting. The release adds support for TCPRoute, UDPRoute, and GRPCRoute alongside existing HTTPRoute and TLSRoute support, covering both on-premises and Konnect-based gateway deployments. It also extends tag propagation from Gateway API resources to Konnect for multi-tenant traffic ownership and cost tracking, while native HorizontalPodAutoscaler support enables managed scaling for AI and Event Gateway data planes.

Kong has introduced a Portal MCP Server that provides each Dev Portal with a governed, AI-native way for agents to discover and read API documentation and specifications without scraping browser-oriented HTML. Enabled through a single per-portal toggle, the server preserves existing authentication, role-based authorization, and content restrictions, ensuring agents access only the same material available to the developers they represent. OAuth 2.0 connections using Authorization Code with PKCE allow tools such as Claude, Cursor, VS Code, and custom applications to connect through standard flows, with controls for trusted clients and approved redirect URLs. The release builds on related Dev Portal capabilities, including Markdown versions of pages for cleaner model consumption, write access through the existing Kong Konnect MCP Server for creating and publishing portal content, and the general availability of the Dev Portal Toolkit VS Code extension for local content previews.

Rising enterprise use of AI models is making token consumption a significant and difficult-to-predict operating cost, as prompt length, model selection, retries, and agent behavior can produce widely varying expenses for similar workloads. The discussion argues that organizations need real-time token cost management combining behavioral optimization, such as routing simpler tasks to less expensive models, with comprehensive governance across models, tools, contextual data sources, providers, clouds, and on-premises environments. Citing Kong’s research, it notes that 99% of organizations lack AI-specific governance controls and that many already use multiple models, increasing the risk of untracked “shadow” token consumption. It identifies ungoverned model choice, delayed spend reporting, uncertain output quality, and siloed visibility as key causes of budget overruns. As a proposed solution, it advocates routing AI traffic through an AI gateway that can attribute costs by user and team, convert token usage into monetary terms, enforce rate limits and budgets, route requests based on cost and intent, and support real-time chargeback or showback, rather than relying solely on retrospective dashboards.

Enterprise AI governance is presented as a composed stack that combines oversight functions such as policy management, inventories, approvals, risk assessments, documentation, and audit evidence with runtime controls that govern live traffic among models, APIs, MCP tools, agents, and event streams. As AI adoption accelerates, organizations are encouraged to align governance programs with frameworks such as NIST and ISO 42001 and organize them around accountability, transparency, risk and security, and operational control. AI gateways can add AI-specific enforcement—including identity controls, access restrictions, quotas, token metering, model routing, prompt-response policies, logging, and sensitive-data handling—while standard API gateways and event gateways govern related API and streaming access. The discussion emphasizes that gateways support but do not replace downstream authorization, broader legal and model-governance processes, or human oversight, particularly for agentic systems that may take consequential actions. Effective programs should address shadow AI, maintain auditable records of decisions and tool calls, distinguish spend reporting from active budget enforcement, and evaluate technology according to coverage, policy enforcement, identity, evidence, regulatory alignment, deployment needs, integrations, and operational reliability.

Insomnia versions 13.2 and 13.3 introduce interface and workflow updates intended to simplify API development and organization. Documents and Collections have been unified into API Collections, allowing OpenAPI specifications and request collections to be kept together while remaining optional, and legacy Unit Tests are now hidden by default in favor of after-response scripts, though they remain supported. Focus Mode lets users narrow the sidebar to a selected API Collection, with an option to retain the broader v13 navigation style, alongside improved drag-and-drop organization. Environment selection and private sub-environment creation have been streamlined across project, collection, and folder levels, with private environments retaining data locally and external vault integrations recommended for sensitive secrets. Additional changes include improved team vault-secret collaboration, clearer local Git project folder names, OAuth2 request logs in the console timeline, and undo/redo fixes, while Konnect integration has moved to the Organization menu.

Kong API Gateway 3.16 introduces runtime and governance capabilities intended to simplify production debugging, consumer-specific policy enforcement, usage controls, and regulatory compliance without downtime or custom plugins. Administrators can dynamically adjust log levels on Hybrid Data Planes, including Konnect and on-premises deployments, with an optional TTL that automatically restores the prior setting. New expressible plugin fields use Common Expression Language expressions to tailor selected ACL and rate-limiting plugin settings per request based on consumer metadata, principal information, or validated token claims. The new OpenMeter-powered Entitlement Enforcement plugin blocks requests when prepaid credits, usage limits, or plan entitlements are exhausted, extending metering into real-time access governance. Kong Gateway Enterprise also offers a separately distributed FIPS 140-3-compliant package, supporting organizations in regulated sectors as NIST transitions away from FIPS 140-2 validation.

Kong has received Great Place To Work Certification in the United States for the fourth time, with 95% of its employees reporting a positive workplace experience compared with 57% at a typical U.S. company. Based on employee responses to the Trust Index survey, high ratings included the ability to take necessary time off, feeling welcomed upon joining, confidence in management’s ethics, authenticity at work, and a sense of individual impact. The company, which has 332 U.S.-based employees within a global San Francisco-headquartered organization, was also ranked No. 30 among small and medium employers on Fortune’s 2026 Best Workplaces in the Bay Area list. Kong attributes its culture to intentional communication, leadership openness, early opportunities for new hires to contribute, recognition programs, and employee feedback that informs processes as the company grows. It positions its work on API and AI traffic governance, security, and connectivity as an opportunity for employees to address emerging technical challenges while balancing rapid growth, operational discipline, and an inclusive workplace culture.

Kong has announced a rebrand intended to reflect its shift from an API connectivity company to an AI connectivity company, emphasizing a platform for securing, governing, accelerating, and monetizing AI and API traffic across models and cloud environments. The new identity features “electric lime” as its primary color, visual references to Renaissance art and the company’s Italian roots, and Karl, a young light-gray gorilla mascot with an electric-lime backpack designed for visibility at small sizes. The rebrand builds on Kong’s history, which began in Milan in 2009 and included the 2015 open-sourcing of Kong Gateway, whose early “Kong Walker” gorilla logo evolved from collaborations involving designer Dave Desandro and engineer Thibault Charbonnier. Kong says its expanding AI Gateway business and the need for AI governance influenced the redesign, while the updated visual system retains elements of its longstanding gorilla imagery and will appear across its website, events, marketing materials, and other channels.

Kong and Straiker have partnered to integrate agent-specific security capabilities with Kong AI Gateway, allowing enterprises to govern AI traffic through Kong while using Straiker to analyze agent behavior, test attack paths, detect risks such as prompt injection and data exfiltration, and return enforcement decisions that Kong applies inline. The integration supports LLM traffic, MCP tool calls, and agent-to-agent communication, preserving Kong as the routing, observability, policy, and enforcement layer rather than requiring a separate gateway architecture. For incidents extending beyond a single unsafe request, Straiker’s Agentic Kill Switch can contain compromised agents by suspending sessions, revoking tool access, freezing memory, or taking individual agents or fleets offline. The companies position the approach for regulated industries and recommend a phased rollout that begins with observation of production traffic, followed by enforcement of validated controls and, where needed, expanded inventory, testing, posture management, and agent-level containment.

Kong announces that it was named a Leader in The Forrester Wave: API Management Software, Q3 2026, with Forrester highlighting its suitability for cloud-native environments and emphasis on AI governance. The company says the report recognizes its strategy to evolve API management into an AI control plane supporting broad protocols, event-driven architectures, gateway modernization, and business alignment, while stating that Kong offers more LLM governance policies than other evaluated vendors. Forrester also cited Kong’s OpenMeter acquisition for customizable analytics, monetization, and LLM token chargeback capabilities, along with growing demand from systems integrator partners and positive customer feedback on support and gateway performance. Kong notes that IDC also named it a Leader in its 2026 API Management assessment, reflecting a broader industry view that API management, governance, and federated hybrid-cloud platforms are increasingly central to secure AI adoption, software development, and connectivity.

Kong Gateway now offers FIPS 140-3–validated builds alongside its existing FIPS 140-2 artifacts, allowing organizations in federally regulated, financial, healthcare, and PCI-governed environments to adopt the newer cryptographic standard without disrupting current deployments. FIPS 140-3, administered through NIST and Canada’s CCCS Cryptographic Module Validation Program, supersedes 140-2 with alignment to ISO/IEC 19790 and stricter requirements for testing, attack mitigation, and lifecycle assurance. Kong’s new build uses the FIPS 140-3–validated OpenSSL 3.1.2 provider, supports TLS 1.3 as well as TLS 1.2, applies a narrower set of modern TLS 1.2 cipher suites, and centralizes FIPS state and cryptographic policy across deployment environments and signing features. The release also rejects weak keys and unsupported EdDSA defaults in FIPS mode, while treating post-quantum TLS as a separate non-FIPS feature that is disabled with a warning when FIPS mode is active. Existing users can remain on FIPS 140-2, while those moving to the new standard can select the dedicated `-fips-140-3` package or image without changing their gateway architecture.

Kong Gateway Enterprise 3.4 reached end of life on August 31, 2026, ending its full support coverage. It will receive 12 months of sunset support through August 2027, primarily to assist customers in upgrading to a currently supported Kong Gateway Enterprise version. Kong advises organizations using version 3.4 to upgrade promptly to maintain support, reduce vulnerabilities, and benefit from improved deployment capabilities and platform visibility.

Kong AI Gateway 2.0 has reached general availability, providing a production-supported runtime, control plane, and Kong Konnect integration for governing AI models, agents, and MCP traffic. New capabilities include MCP Server Bundling, which combines multiple MCP servers behind one route while using access-control policies to expose only authorized tools to each caller; modality-aware pricing that tracks costs across text, audio, images, video, and cache operations; and expanded native support for Kimi, Microsoft Foundry, Amazon SageMaker, and AWS Bedrock AgentCore through IAM-based SigV4 authentication. The release also integrates Kong Identity principals into AI Gateway policies, enabling organization-wide identity-based rate limiting, access controls, analytics, and cost attribution in addition to existing Consumer-based configurations. Existing AI Gateway plugin users can continue operating on Kong Gateway 3.x, with AI plugins supported through the 3.14 LTS release, while Kong recommends migration before version 3.18 using kongctl, documentation, or assisted migration services.