Kong Unveils Volcano Platform for Building and Running AI Agents

Kong Inc. on September 30, 2026 announced Volcano, an AI-native platform that combines durable compute, branchable PostgreSQL databases, file storage, user authentication and real-time services into a single foundation for building, deploying and operating AI agents and modern web applications.

Kong said a production agent requires more than a model call: it needs compute that executes durably, a database that remembers, authentication for safe interaction with users, storage for files, workflows that coordinate tasks and observability into what happens once it is running. The release states that most developers assemble those pieces from separate providers, each with its own configuration, billing and failure modes, and that Volcano combines them in one platform so teams do not have to piece that infrastructure together themselves.

“Volcano allows us to build in the AI era with a platform that is truly ‘batteries included,'” said Marco Palladino, Kong’s CTO and co-founder, in the announcement. “Developers shouldn’t have to stitch together infrastructure to turn an AI agent into a production application.”

Volcano ships with first-class integrations with the agentic coding tools Claude Code, OpenAI Codex and Cursor, and Palladino said a developer can take an agent from zero to production in a single prompt. Kong noted that the Volcano announced is a new product, separate from the VolcanoSDK it announced in 2025.

Data, Compute and Durable Execution

The release lists branchable PostgreSQL databases provisioned in seconds, with built-in backups and vector support for AI workloads. Volcano’s database documentation specifies PostgreSQL 15 or 16 (16 by default), with copy-on-write branches that fork a database’s schema, rows, roles and policies behind a separate connection string and are usually ready in under a minute. Branch lifetimes run from one hour to 30 days, each database supports up to 10 branches on the Hobby plan and 25 on Superagent, and pgvector installs through CREATE EXTENSION for embedding workloads. Databases accept direct connections on port 5432 over TLS, enforce row-level security through functions such as auth.uid() and auth.role(), and come in six compute tiers from volcano-db-xs to volcano-db-2xl. Backups and a seven-day point-in-time restore window are available only on the Superagent plan.

For long-running work, the release lists durable workflows and jobs with built-in leader election. Volcano’s durable functions documentation describes checkpointing at every step, so an execution that crashes or waits resumes from the last completed step, and a single execution can run for up to 366 days. Handlers are written in JavaScript on Node.js 22 or 24, or in Python 3.13 or 3.14. Each step is limited to 300 seconds on Hobby and 900 seconds on Superagent, and the two plans allow 10 and 100 concurrent executions respectively. Steps that fail retry with exponential backoff, an atMostOnce setting records a step’s result before execution so the step never runs twice, cron schedules and idempotent starts are built in, and a workflow can pause for human approval through ctx.waitUntil, with no compute charged while an execution waits.

Functions, Hosting and Application Services

The release lists Edge Functions that run code and AI agents across multiple regions and languages with autoscaling and geofencing for data-sensitive workloads; Volcano’s product site specifies Node.js, Python or Ruby served from the closest of eight regions. Frontend hosting deploys Next.js applications from source to a global CDN with TLS and DDoS protection, and a failed build never displaces the version currently live.

Authentication is offered as a service with no cap on users or monthly active users, plus third-party SSO support and built-in anonymous signups; the product site lists Google, GitHub, Microsoft and Apple among its sign-in options. Real-time services listen for database events and provide messaging, broadcast and presence tracking, which Kong says can coordinate multiple agents in real time. File storage is private by default, with SQL-style access policies, optional public links for individual files and uploads that can resume for large files. Distributed locks coordinate agents, workers and durable functions so that only one holder works on a task at a time, with leases that renew for the duration of the work, and the release lists locks, queues and optional human approvals for building single- and multi-crew agents.

Availability and Same-Day Konnect Launches

The Volcano site offers a free Hobby account and distributes its command-line tool through npm. Volcano is compatible with the broader Kong Konnect platform for enterprise governance of AI agents and AI connectivity at scale, and Kong said it will keep expanding its cloud services for developers and AI builders, with additional capabilities planned across compute, data, application services and agent infrastructure. Availability for sandboxing compute will follow shortly after the announcement, the company said.

At its API & AI Summit, also on September 30, 2026, Kong announced the evolution of Kong Konnect into what it calls the AI Connectivity Platform, according to Kong’s summit recap. The recap frames a production agent’s path as six stages, from reacting to events and discovering the tools, models and other agents it is sanctioned to use, through getting credentialed, consuming context efficiently and paying for itself, to being understood through queryable, traceable and debuggable operation. Launches mapped to those stages include Webhook Engine in private beta, Konnect Catalog and Context Mesh at general availability, Token Vault and an Agent & MCP Registry coming soon, and AI cost management and Advanced AI Observability in early access for design partners.