{"slug": "kong-volcano-one-platform-to-replace-your-entire-agent-stack", "title": "Kong Volcano: One Platform to Replace Your Entire Agent Stack", "summary": "Kong shipped Volcano, a full-stack agent infrastructure platform bundling eight generally available services — including durable compute with 366-day execution windows, branchable PostgreSQL with pgvector, edge functions across eight regions, and MCP-native agent hosting — on September 30, alongside generally available Context Mesh and Kong AI Gateway 2.2. Kong said Context Mesh's Code Mode cuts starting context by 69 percent by letting agents retrieve tool schemas on demand from existing Kong Konnect APIs, while AI Gateway 2.2 adds MCP tool governance, identity-aware rate limiting and cost attribution, and modality-aware pricing. None of the eight Volcano services shipped with observability at launch; Kong's Advanced AI Observability feature remains on the roadmap.", "body_md": "Kong — the API gateway company behind 350 million open-source downloads — shipped a full-stack agent infrastructure platform on September 30, and it is worth your attention. Volcano bundles branchable Postgres, durable compute with 366-day execution windows, edge functions, auth, file storage, and MCP-native agent hosting into a single platform. The pitch is blunt: stop assembling five services to run one agent.\n\n## This Is Not the 2025 SDK\n\nDevelopers familiar with Kong will know the [Volcano Agent SDK](https://github.com/Kong/volcano-agent-sdk) — the TypeScript library that lets you build a working agent in about nine lines of code. That is still open source and free. The Volcano announced at Kong’s [API + AI Summit](https://konghq.com/blog/news/ai-summit-2026-launch-recap) is a different product entirely: a hosted platform that competes directly with the stack most agent developers have cobbled together.\n\nThat stack usually looks like Supabase for Postgres and auth, Temporal for durable workflows, Fly.io for edge deployment, and something else for file storage and real-time coordination. Volcano replaces all of it with one bill and one CLI. Whether that trade-off is worth it depends on where you are, but the scope of what Kong shipped deserves a straight look.\n\n## What Is Actually in the Box\n\nVolcano ships eight services at general availability:\n\n- **Durable compute** — Workflows with 366-day execution windows and checkpointing. Agents that fail midway resume from their last checkpoint rather than starting over.\n- **Branchable PostgreSQL** — Copy-on-write database branches that fork schema, rows, roles, and policies in under a minute. The Hobby plan includes 10 branches; Superagent includes 25. pgvector is built in for embedding workloads.\n- **Edge Functions** — Eight regions, autoscaling, and geofencing for workloads with data-residency requirements.\n- **Real-time coordination** — Locks, queues, multi-agent messaging, and optional human-in-the-loop approval gates.\n- **Auth, SSO, file storage, and frontend hosting** — The rest of the application layer, included.\n\nThe branchable Postgres is the most technically interesting piece. Supabase does not offer this natively. Being able to fork your production database schema into a short-lived branch for an agent test run, then discard it cleanly, is genuinely useful and currently rare in managed Postgres offerings.\n\n## Context Mesh: The Enterprise Angle\n\nAlongside Volcano, Kong shipped [Context Mesh](https://developer.konghq.com/context-mesh/) as generally available. Context Mesh takes your existing APIs — everything already flowing through Kong Konnect — and exposes them to agents as a single, governed MCP server. Standard MCP tool calling does not scale when your API surface has hundreds of operations; exposing all of them to a model’s context window is expensive and slow.\n\nContext Mesh solves this with Code Mode: agents search the available API surface, retrieve the relevant tool schema on demand, and execute it in one round trip. Documented benchmarks show a 69 percent reduction in starting context. For enterprises already running Kong Gateway, this is the lowest-friction path to making existing systems agent-accessible — no rewrite required, policies and access controls inherited automatically.\n\n[Kong AI Gateway 2.2](https://konghq.com/blog/product-releases/kong-ai-gateway-2-2), also generally available on September 30, adds MCP tool governance (consolidate multiple servers into one route, surface only what each caller is authorized to see), identity-aware rate limiting and cost attribution, and modality-aware pricing that tracks text, audio, image, and video spend separately.\n\n## The Part Kong Did Not Mention Loudly\n\nNone of the eight Volcano services shipped observability at launch. Agents can be deployed into production before their developer can trace what they actually did across a full session. Kong’s Advanced AI Observability feature exists on the roadmap but is not generally available.\n\nCTO Marco Palladino described Volcano as going “from zero to running an agent in production in one prompt.” That is accurate. It is also possible to run that agent with no visibility into how it behaved once it was there. For teams with audit requirements or agents operating on sensitive data, that gap matters more than the convenience of a one-prompt deploy.\n\n## SDK or Platform — You Do Not Have to Choose Today\n\nThe [Volcano Agent SDK](https://github.com/Kong/volcano-agent-sdk) is MIT-licensed and works without the Volcano platform. If you want the MCP-first agent framework without committing to Kong’s hosted infrastructure, the SDK is a legitimate standalone option. OpenTelemetry traces export to Jaeger, Prometheus, or DataDog if you already have observability covered.\n\nThe hosted platform makes sense if you are building a new agent-powered product and want to skip the infrastructure assembly phase entirely, or if you are an enterprise team already on Kong Konnect where Context Mesh plugs in without rearchitecting. It makes less sense if your existing Supabase and Temporal setup already works and observability is a hard requirement today.\n\n## The Bigger Signal\n\nKong spent 15 years building trust as the layer between your services and the outside world. Volcano is a bet that the same trust extends to hosting the agents themselves. An API gateway company pivoting to agent infrastructure is a meaningful market signal — and [Context Mesh](https://developer.konghq.com/context-mesh/), already GA and already useful without the full platform, is the most practical thing to evaluate right now regardless of how deep you want to go.", "url": "https://wpnews.pro/news/kong-volcano-one-platform-to-replace-your-entire-agent-stack", "canonical_source": "https://byteiota.com/kong-volcano-agent-platform/", "published_at": "2026-10-17 06:07:57+00:00", "updated_at": "2026-10-17 07:03:54.793774+00:00", "lang": "en", "topics": ["ai-agents", "agent-protocols", "ai-infrastructure", "ai-tools", "developer-tools"], "entities": ["Kong", "Volcano", "Volcano Agent SDK", "Context Mesh", "Kong AI Gateway 2.2", "Kong Konnect", "Supabase", "Temporal"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/kong-volcano-one-platform-to-replace-your-entire-agent-stack", "markdown": "https://wpnews.pro/news/kong-volcano-one-platform-to-replace-your-entire-agent-stack.md", "text": "https://wpnews.pro/news/kong-volcano-one-platform-to-replace-your-entire-agent-stack.txt", "jsonld": "https://wpnews.pro/news/kong-volcano-one-platform-to-replace-your-entire-agent-stack.jsonld"}}