Closed preview · Own infrastructure
One environment where your team ships, partners collaborate, and agents operate — conversation, intelligence, code, and infrastructure as one system.
@agent set up the backend for the places app: GraphQL API, users and places with sessions, activity, and notifications, media uploads behind a public CDN route, place search, embeddings, and a recommendation model trained on activity. Deploy it to staging.
Planning stack-42: 2 functions, 2 workers, 6 tables, 2 search
indexes, 2 inference endpoints, 1 training job, 3 buckets, 4 routes. Deploying to staging.
Deployed. 22 resources live in eu-de-1. GraphQL on app-api, media
on media-public-cdn, first training run started. No console opened, no keys issued.
Shipping one product now means a chat tool, a ticket board, a repo, a CI pipeline, a preview host, a few AI subscriptions, and a dozen cloud services — each with its own access model and its own invoice. Every partner needs a seat in some of them. Every agent needs keys to most of them.
~20 tools, ~20 access models, five or more invoices. Assembled by you, inherited by everyone you let in.
stack.json, previews. The whole application as
one stack, deployed with one command.
One environment. One identity model. One ceiling for partners and agents.
Compute, data, and AI as one system — not a collage of managed services with separate APIs, billing, and operational models.
APIs, web apps, and long-running jobs in one deployable stack. Go, Node.js, Python, and PHP. Auto-scaling execution without capacity planning or idle fleets.
High-performance NoSQL datastore, serverless by default — no clusters to size, no provisioning cycles.
Object storage in the same stack as compute and data — for media, logs, worker volumes, and direct uploads via pre-signed URLs.
Full-text and vector indexing on your tables and objects — native to the stack, not a third-party sidecar.
Serverless inference endpoints and AutoML training on your own tables, on the same fabric as your data — bring your own model, or call frontier models through the same gateway. No separate AI platform to stitch in.
Routes, domains, API keys, CDN edge, secrets, and alerts — the operational surface you should not have to assemble yourself.
Who builds software is changing, and intelligence has become a commodity. Both point at the same design: one system that people and agents share — intelligence plugged in from outside, control kept inside.
A few people with agents now ship what used to take a department, and coding agents write most of the glue. Hand them thirty services and thirty IAM policies and they inherit your cloud. Hand them one environment with a ceiling and they become useful without becoming dangerous. The runtime and the agent have to be the same system.
Frontier models improve every quarter and are interchangeable through agent harnesses. What is scarce is an environment where they can act on real systems — code, infrastructure, data, and the conversation with the customer — with context, permissions, and a ceiling. That environment is the product.
Not a wrapper around Lambda, DynamoDB, and S3. One control plane schedules compute, data, search, inference, and storage on infrastructure we run ourselves — with a pluggable intelligence layer.
We don't rent capacity from a hyperscaler and pass the bill on. Owning the stack from hardware to API is what lets us offer pay-for-what-runs and scale-to-zero without a catch. The same logic applies to models: we don't lock you into one lab.
Hardware we operate, not resold capacity. No hyperscaler underneath, no pass-through pricing, no vendor whose roadmap the fabric depends on.
A single system places and scales functions, workers, tables, search, inference, and buckets. One deploy, one identity model, one meter.
Frontier models from Anthropic, OpenAI, and Google, and harnesses like Cursor SDK, Claude Code, and Codex plug in from outside through the same gateway and the same sandbox ceiling. Hosted, open, and AutoML models run on the fabric next to your data. We own the parts that should never change hands and swap the parts that change every quarter.
Scale-to-zero, healing, and placement are the platform's job. No capacity planning, no idle fleets, no on-call for the fabric.
Bring your own hardware into the same fabric. On-premise and hybrid deployments run the same stack with the same control plane — for data that has to stay where it is, without a second platform.
Conversation, intelligence, code, and infrastructure are one system. Your team deploys the stack. Partners join the project. Agents work on the repo, in the conversation, and against the real environment. Same system, different ceilings — nobody inherits the cloud.
Can the client see the new pricing page before the 2pm review? They don't have console access.
@agent build the pricing-page branch
and put a preview on the client route.
Preview is live on client-preview, built from
pricing-page. Public route, no console, no keys. It scales to zero after the review.
Perfect — forwarding it to the client now.
Declare functions, tables, storage, routes, and apps as one unit — via CLI, API, or Console. Environments and access keys stay with your organization.
Clients, vendors, and guests join the project’s conversations — in the Console, a Telegram group, or via WhatsApp — without infrastructure, credentials, or environments. Every partner sees the product before they see the platform. That is the distribution.
Agents clone, change, and preview against the live environment inside a sandbox with a ceiling. You keep control of what merges and what runs. They use the environment; they do not become a separate platform.
Monthly plans with credits included, credit packs on demand, and enterprise agreements for regulated workloads. One meter across compute, storage, network, search, training, inference, and agent runs; scale-to-zero removes idle spend instead of papering over it with reserved capacity.
For serious solo builders and studios shipping with partners and agents.
For development agencies: client workspaces, white-label previews, priority support.
For ISVs and regulated teams: reserved capacity, SLA, dedicated zone.
Custom volumes, dedicated concurrency, contractual SLAs.
Preview pricing. Not final. See full pricing and credit packs →
KERNSTACK is developed by coding.berlin. The control plane, runtimes, datastore, search, inference, gateway, and console are built in-house — there is no hyperscaler underneath and no vendor whose roadmap we depend on. The platform runs on our own infrastructure in Europe and is in closed preview. For teams with data-residency requirements, the same stack runs on-premise via User Nodes.
KERNSTACK is a cloud platform built from scratch as one system: compute, data, search, inference, and storage on a single fabric, running on our own infrastructure in Europe. Your team owns the stack; partners and agents work in that environment without taking the keys. One credit balance; pay only for what runs.
Hyperscalers evolved from VM primitives and expose hundreds of loosely coupled services. KERNSTACK is one control plane for apps, data, and AI — and one environment for the people and agents who work on them. You do not assemble a cloud, then bolt on a workspace and an agent tool. The runtime and the collaborators are the same system.
No. KERNSTACK runs on dedicated bare-metal servers we operate in Europe. We are not wrapping Lambda, DynamoDB, or S3 behind another brand. The same system supports public, private, and hybrid deployments, including on-premise User Nodes — your hardware, joined to the same fabric — when data locality or sovereignty requires it.
No. Partners join the project as collaborators: the board, the conversations, the work. Agents operate in a sandbox with an explicit ceiling. Environments, access keys, and deployed resources stay with your team. That ceiling is the product, not a setting you have to remember.
KERNSTACK is in closed preview. Reach out for access. Preview environments may change — APIs, features, and pricing are not final, and we do not yet offer production SLAs.
Closed preview. Talk to us about access or a design partnership.