Changelog · 2026
The customer-facing features and improvements we released each month, from January to November, most important first.
01
Debug with AI can now read your app's GitHub repository at the exact commit that failed to build or is running now, and it shows each file it reads. It finds the cause of a broken build in your Dockerfile, package files or config instead of asking you to paste them. Fixing a failed deploy takes one chat, and only people who can edit the app get repository access.
01
Team accounts are here: projects, apps, Grids, servers and billing belong to your organization, with members, roles, invitations and an audit log. Each division can run its own Grids while IT keeps one view of everything. Onboarding and offboarding take seconds.
02
Organizations can turn on SSO with their identity provider, verify their domains and require SSO for every member. Your team signs in the way it already does at work. Access is controlled centrally and removed automatically when someone leaves.
03
Roll a GitHub-deployed app back to any earlier build instantly, without rebuilding. A bad release is undone in seconds. Ship faster, because mistakes are cheap to fix.
04
A redesigned Grid page shows every policy at a glance: GPU limits, grace periods, per-container caps, idle reclaim and fair share. Limits are applied live to running containers. Owners set the rules once, and Kinesis enforces them.
05
Launch an app on a Grid before its machines arrive, and it starts as soon as machines that fit it join. Spot machines are leased only while a running app needs them, and a new "Apps on this grid" view shows which apps fit which machines. Owners stop paying for GPUs no app uses, and teams are never stuck behind a hard block.
06
Redesigned Servers and Grids pages show each machine's hardware, live CPU and GPU usage, whose apps run on it and who added it, with a summary of the whole fleet on top. Both pages load in milliseconds, even for large fleets, and keep a full audit trail. IT gets clear visibility across the whole organization, and capacity planning and cost conversations start from facts.
07
Apps can now run up to 10,000 replicas, up from 20, for large batch jobs and high-traffic services. A new spread group keeps apps with the same group name on different machines, so one machine going down never takes out every copy of a database or cache. Teams scale out without giving up resilience.
08
Updates now pull the new image before they stop the old container, and each app can set its own shutdown grace period. The Deployments tab shows a new build as Deploying until your servers really run it, so Active always means live. Updates cause less downtime, and teams always know which build is serving.
09
Spot machines leased from the marketplace no longer bill for the time they are offline. When a leased machine goes dark, the Grid lets it go and leases a fresh one, so apps keep running. Customers pay only for compute they can actually use.
10
Every internal API call now checks the caller's role, servers verify the control plane's identity before they connect, and machine keys are created on the machine itself. These changes close whole classes of attack before anyone can try them. Customers get stronger protection without changing anything.
01
Everything you do in the portal can now be automated through a documented REST API (OpenAPI spec included) with secure API keys. Plug Kinesis into your CI/CD, internal tools and scripts. Infrastructure as code, without the infrastructure.
02
Kinesis now has an MCP server with secure OAuth sign-in, so AI assistants and agents can deploy and manage your apps, Grids and builds. Ask your assistant to ship a model, and it does. This is the shortest path from idea to running AI app.
03
Grid owners can cap how many apps, instances and GPUs each user can take. Users see clear notices when they hit a limit. Shared capacity stays fair and predictable for every team.
04
Adding a server now starts with where it is connected, and recommends the right method, including Kinesis Remote Reach for machines behind a firewall or NAT. You don't need to open inbound ports. Office PCs, labs and on-prem servers join in minutes, securely.
05
Cloud servers now run a security monitor by default that detects suspicious container activity, stops it and reports it to you. Threats get caught as they happen, not after the damage. Peace of mind comes standard.
06
Start a new app from a template that already has the right image, ports and resources. Common workloads go live without reading any docs. New users reach their first running app faster.
07
App, server and log pages now show placement checks, work orders and Grid state in one place. When an app waits for a machine, you see exactly which rule holds it back. Problems that used to need a support ticket now take a glance.
08
Spark, the Kinesis desktop app for Windows, now links a PC to your account in a few clicks and lets you pause it or set a schedule. Office and lab PCs add compute during the hours they would sit idle. Owners stay in control of when their machine works.
09
Kinesis now flags environment variables that look like secrets and hides their values across listings and APIs. Passwords and API keys don't leak through screenshots or shared screens. Security best practice becomes the default.
10
Spot instances now include GeForce RTX 20-series and newer cards and workstation RTX GPUs, named by their full model. Customers get more choice at more price points. Smaller workloads find affordable hardware that fits.
01
Grid apps can now keep up to 4 persistent storage volumes that survive restarts and redeploys. Databases, model caches and user uploads stay safe. Stateful workloads now run on Kinesis as easily as stateless ones.
02
Split data-center GPUs (A100 and H100 class) into isolated slices with a visual layout designer, and give each app its own slice. One powerful GPU can now serve many smaller workloads safely. Do more with less hardware.
03
Grids can now reclaim GPUs from apps that sit idle, with configurable thresholds and a clear warning before anything is stopped. Scarce GPUs go back to the people who are actively using them. Every GPU you pay for does real work.
04
When a Grid is full, apps now wait in a fair queue and take turns instead of first-come-first-served. No single user can hog shared capacity. Every team gets its share, automatically.
05
Share Grids with Editor and User roles, see live CPU, RAM, GPU and VRAM use for every app, and stop or restart apps from one screen. Grid owners get full visibility and control over shared capacity. IT stays in charge without getting in the way.
06
If a GPU is out of stock when a hibernated machine wakes, Kinesis keeps trying and shows you the status, and owners get an alert if a machine fails to hibernate. Nothing gets stuck silently. You can trust hibernation to save money without risking your capacity.
07
The App Gallery now deploys to your Grid by default when you have one. Models and apps land on the hardware you already pay for, with no extra setup. Getting value from your own compute takes one click.
01
Grids now combine your own machines, on-demand cloud machines and rented Spot Instances from the Kinesis marketplace in one pool. Grow and shrink compute in minutes, at the price that fits each workload. One platform, one view, for every kind of compute you use.
02
Upload your SSH keys once and assign them to your cloud and BYOC machines for direct access. Engineers get into their machines with the tools they already use. Your keys, your control.
03
Rank machines by priority inside a Grid, and let one app run across several Grids. Your preferred hardware takes the work first, and the rest of the fleet is there for overflow. Smarter placement means better performance and lower cost.
04
One app can now serve several custom domains, including root (apex) domains like yourcompany.com. Run regional sites, brand variants or migrations from a single deployment. Every domain gets automatic SSL.
05
Approved suppliers can list whole machines on the Kinesis marketplace and track their earnings in the new Provider Hub. Hardware that would sit unused now pays for itself. More supply on the network also means more choice for everyone who rents.
06
A100 data-center GPUs can now be provisioned straight from the Grids page. Large training and inference jobs get top-tier hardware in minutes, without a long-term contract. You scale up for the big job and give it back when you are done.
07
Apps that run on a Grid shared with you now use that Grid's machines without extra per-use charges or balance checks. The Grid owner pays for the capacity once, and every team uses it freely. Sharing compute inside a company is finally as simple as sharing a folder.
01
An AI assistant now lives in the portal. It can read your build failures and app logs, explain what went wrong and suggest Dockerfile fixes. Help is instant, in context, and available around the clock, with no ticket queue.
02
Machines being provisioned now show live progress, instance details, and Retry or Dismiss options if something fails. You always know what is happening and what to do next. No more guessing whether capacity is coming.
03
Apps now show precise states as they change, including a new "Start Blocked" state that explains why an app can't start. Small changes to a running app no longer get blocked for no reason. Less confusion, faster fixes.
04
Cloud machines can now use fast local SSD for images and containers, and built-in instance storage is no longer billed separately. Big models load faster and costs go down. Better performance for less money.
05
Pin any machine so it never hibernates automatically. Always-on services stay ready, while the rest of your fleet keeps saving money. You get both savings and reliability, machine by machine.
06
Traffic to your apps now runs through a rebuilt App Proxy layer that checks its own health and repairs routes and certificates automatically. Your endpoints stay up even when part of the network has a bad day. Reliability improves without any change on your side.
07
A new list view for Grids and servers adds search, pagination and a per-app view. Teams with hundreds of machines find the one they need in seconds. The portal scales with your fleet.
01
Invite collaborators to any project or app with the right role, and keep everyone up to date with in-app and email notifications. Teams work together in one place instead of sharing logins. Access is clear, and you can remove it at any time.
02
Protect your account with an authenticator app or a passkey/security key, with backup codes and trusted devices. This keeps your workloads and billing safe even if a password leaks. It is enterprise-grade security that takes seconds to turn on.
03
Upload a Docker image file straight to Kinesis and deploy it. No registry needed. Air-gapped builds, private models and teams without a registry can now ship with one upload.
04
Bring your own compute with a single install command and a personal token that you can regenerate at any time. Adding a server takes a minute, and access stays under your control. Scaling from one machine to a hundred is as simple as running one command.
05
Every project and app now has an activity trail that shows who changed what, and when. Teams can trace any change in seconds. Audits and post-mortems become simple.
06
Pay by card, wire or ACH, get invoices that list every machine, and download all your charges and credits as CSV. Finance teams get the documents they need without asking support. Kinesis now fits the way larger companies buy.
07
Stream logs from several instances of an app at once, in tabs you arrange yourself. Problems that only show up on one machine are easy to spot. Debugging scaled-out apps no longer means jumping between pages.
01
See live GPU and CPU availability before you provision, with "high demand" labels. When a machine is out of stock, you can order it anyway, and Kinesis gets it for you as soon as it is free. No more refreshing pages to catch an open GPU.
02
App setup now has clear Execution and Compute strategy tabs with card-style choices. You describe how your app should run, and Kinesis handles the hardware details. New users get to a working deploy faster, and experts still get full control.
03
See how long your balance will keep your machines running, protect critical machines from automatic shutdown, and get email warnings early. No surprise outages because a balance ran low. Planning ahead becomes easy.
04
Machines that hit a critical failure now reboot and recover on their own, and you can replace a faulty machine with a fresh one in one click. Your apps spend less time down, and your team spends less time on support. Reliability without babysitting.
05
Choose how long logs are kept, download or delete them, browse history, and view a machine's system logs from the portal. Debugging takes minutes instead of SSH sessions. You decide how much data you keep, for cost and for compliance.
06
The App Gallery added the latest open models, including Gemma 4, Qwen 3.5 and Qwen 3.6. Teams try new models the week they come out, with one click. Staying current with open-source AI takes no extra engineering.
07
Deploy from private Docker registries with credentials that Kinesis checks before it saves your changes. A wrong password shows up in the form, not as a broken app. Private images become as easy to ship as public ones.
01
Grids now provision AWS, Azure and Kinesis cloud machines on demand, straight from the portal, with GPU and CPU options side by side. No separate cloud accounts or consoles to manage. Your own hardware and on-demand cloud capacity live in one place.
02
Attach your own domain to any app, and Kinesis issues and manages the SSL certificate for you. Your users see your brand, not ours. No certificate chores, ever.
03
Grid owners can now share their machines with other users and teams. An IT admin can buy capacity once and give every team access to it. Expensive GPUs get used by everyone who needs them, not locked to one account.
04
Put machines to sleep, wake them up or reboot them with one click, with clear status at every step. You decide when to save money and when to have capacity ready. Simple controls keep you in charge without SSH or scripts.
05
Dedicated machines now show their price up front, roll into a fleet cost overview and appear on the same invoices as everything else. Finance teams see one clear bill instead of several. You always know what your fleet costs before the invoice arrives.
06
Kinesis now notices when a Docker tag points to a new image and offers to pull the latest version. You always know whether your app runs the code you just pushed. No more redeploys that quietly keep the old build.
07
The app form now suggests ports from your image, lets you override its entrypoint and set shared memory, and keeps shell-based images running. Images that used to need a custom wrapper now run as they are. More of the software you already have deploys on the first try.
08
Pick the storage size for every cloud machine you provision, on AWS, Azure and Kinesis cloud. Big models and datasets fit without workarounds. You pay for the disk you need, not a fixed default.
01
Connect a GitHub repo, pick a branch, and Kinesis builds and deploys it for you. No Docker image needed, with live build logs and a Deployments history. Every push to your branch rebuilds and redeploys automatically, so you get CI/CD without writing a pipeline.
02
No Dockerfile in your repo? Kinesis detects what your project needs and an AI writes the Dockerfile for you, streaming it live as it goes. This removes the most common blocker for teams new to containers. More of your code reaches production, with less DevOps knowledge needed.
03
Hibernated machines now wake up automatically when an app needs them. You get the savings of hibernation with none of the hassle of turning machines back on by hand. Compute is there when your workload asks for it, and asleep when it doesn't.
04
A new visual theme and redesigned Projects and Apps pages show status at a glance, with project search built in. Teams with many apps find what they need faster. Day-to-day management takes fewer clicks.
05
Auto top-up now protects itself: after repeated failed card charges, it turns off and tells you instead of looping. Customers never get stuck in a cycle of failed payments. Billing stays predictable and transparent.
06
Kinesis cloud added new datacenter partners, with storage attached to machines automatically. Customers get more locations and more GPU options to choose from. More supply means better availability and better prices.
07
Every GitHub build now keeps its full log, so you can read it during the build and long after. When a build fails, the reason is one click away. Teams fix broken builds in minutes instead of rerunning them to find out what happened.
01
Anyone can now browse the Kinesis App Gallery before signing up, and pick the exact Ollama model to run with one click. Teams can see what they can run before they commit to anything. Going from "I want to try this model" to a running endpoint takes minutes, not a weekend of setup.
02
The new "Task Completion" restart policy lets batch jobs, data pipelines and one-off tasks run once and stop cleanly when they finish. Before, a finished job could restart again and again, which wasted compute. Now you pay for the work, and the job stops when the work is done.
03
Dedicated machines with no running apps now hibernate on their own. Customers no longer pay for GPUs that sit idle overnight or between projects. The machine stays yours, ready to come back when you need it.
04
Export an app's configuration, import it into a new app, or duplicate an existing one with one click. Moving a setup from staging to production, or sharing it with a teammate, no longer means rebuilding it field by field. The exported file even lists every allowed value, so edits are easy and safe.
05
Environment variables now support placeholders that Kinesis fills in for you, such as your app's own domain, plus a faster bulk-edit mode. You configure an app once, and it stays correct when its address changes. Fewer manual edits mean fewer broken deploys.
06
Apps behind the Kinesis load balancer can now serve gRPC over TLS, and web apps keep each visitor on the same instance with sticky sessions. Modern APIs and stateful web apps scale out without code changes. Your users get a fast, consistent experience no matter how many instances you run.
07
The portal now validates your Docker image before launch and sizes storage from the image itself. Mistakes surface in the form, not as a failed deploy. Fewer surprises, and less time spent reading error logs.