> For the complete documentation index, see [llms.txt](https://docs.ai.neevcloud.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.ai.neevcloud.com/release-notes/nov-2026-release.md).

# November 2026 Release

**Release Type:** Major Feature Release\
**Release Version:** v0.8.1 – v0.9.0

***

## Overview

The headline this release is networking for the VM Service: VMs get a network of their own, with VPCs, subnets, Reserved IPs, NAT gateways and security groups that you control. Alongside it, Agents catch up with Sandboxes — popular open-source agents now come preconfigured against the NeevCloud Model API, and agents pause, resume, fork and roll back like sandboxes do. Sandboxes come with a free monthly allowance and new templates.

* **Your own network for VMs** — VPCs, subnets, Reserved IPs, NAT gateways and security groups
* **Savings Plans for VMs** — commit to a term, without recreating the VM
* **Stop and start GPU Pods** — release the GPUs when you are not using them, and keep your configuration and persistent data
* **Agents that work out of the box** — OpenClaw, Claude Code, Codex CLI, opencode and Hermes, preconfigured against the NeevCloud Model API
* **Agents that resume, fork and roll back**, with a clearer Agents console
* **1,500 free sandbox minutes every month**, for every verified account, no card required
* **Paused sandboxes and agents wake up on their own** on the first request — no separate resume call
* **File-based snapshots**, **Docker** and **Playwright** templates, labels and resumable uploads for sandboxes
* **More dependable snapshots and restores** — you get your state back, or a clear error
* **Connect Claude Code and other MCP clients straight to your sandboxes** through a single MCP endpoint
* **Build your own agent on a sandbox** — the NeevCloud SDKs 0.9.0 ship sandbox tools with adapters for the OpenAI Agents SDK, Claude Agent SDK, Vercel AI SDK and LangGraph
* **`/v1/responses`** and **`/anthropic/v1/messages`** on AI Inference, alongside chat completions
* **Business KYC**, alongside the individual flows

Together, these updates give you more control over your network and your compute, and make Agentic Studio easier to adopt and run.

***

## What's New

### GPU AI Service & VM Service

**Your own network for VMs (New).** VMs can now run inside a private network that you control.

* **VPCs and subnets**, with each VM created in a subnet and given a **private IP**
* **A default VPC** in every project and region, so you can get started without designing a network first
* **Reserved IPs** to give a VM a stable public address
* **NAT gateways** that you manage, for outbound access from private subnets
* **Project-level security groups** to control traffic to your VMs
* Network resources are subject to project quotas

See [VM Networking](/vm-service/networking/overview.md), [VPCs and Subnets](/vm-service/networking/vpcs-and-subnets.md), [Security Groups](/vm-service/networking/security-groups.md) and [Reserved IPs](/vm-service/networking/reserved-ips.md).

> Enabling VM networking requires a scheduled maintenance window with brief downtime for running VMs. Your NeevCloud account team will confirm the timing with you before it goes ahead.

**Savings Plans for VMs.** If you intend to keep a VM running for a while, you can now commit to a term — without recreating the VM.

* **Upgrade a running On-Demand VM to a Savings Plan** at any time. Choose 1, 3, 6 or 12 months for a GPU VM, or 3, 6 or 12 months for a CPU VM
* **No interruption.** The plan applies immediately, and your VM keeps running with the same IP addresses, ports, installed software and SSH access
* **Extend an active plan** by a further term, from the same choices. The extension begins the day after your current plan ends, so you lose none of the time left on your current plan
* **See where your plan stands** from the VM list and the VM details page — plan name, start and end dates, and days remaining
* **Expiry warnings.** When 7 or fewer days remain, the VM details page and the VM list both flag it, so a plan does not lapse without you noticing. If a plan does expire, the VM simply reverts to On-Demand

Note that an active plan cannot be cancelled early, and you cannot switch directly from one active plan to another. See [VM Savings Plan](/vm-service/vm-savings-plan.md).

**Stop and start GPU Pods (New).** You can now stop a running GPU Pod when you do not need its GPUs, and start it again later from its details page — without rebuilding your environment.

* **What is kept:** data on the persistent Network Storage volume, the Pod's configuration, SSH keys, and custom TCP ports and UDP port ranges
* **What is released:** the GPU and Pod resources. Container disk data — everything outside the persistent volume — is cleared, so keep anything you need on the volume
* **A confirmation dialog** shows exactly what will be retained and what will be released before the Pod stops
* **Start it again** from the Pod's details page; once it is **Running**, your access options are available as before

Volume storage continues to be charged while a Pod is stopped. If you no longer need the environment, delete the Pod instead. See [Stopping a GPU Pod](/gpu-ai-service/deploying-gpu.md#stopping-a-gpu-pod).

**Also in VM Service:**

* **A redesigned VM create flow**. See [VM Deployment](/vm-service/vm-deployment.md)
* **GPU Pods and GPU VMs draw from the same pool of cards** in a region. Any free card can be launched as either, so capacity is no longer stranded on the side you are not using, and availability reflects what you can actually launch
* **GPU VMs recover correctly from out-of-memory conditions**

> GPU VMs start from a clean OS image, so you install the frameworks and dependencies your workload needs. If you want a ready-made AI environment, [GPU AI Service templates](/gpu-ai-service/ai-templates.md) provide one.

***

### Agentic Studio — Sandboxes

**A free tier.** Every verified account now gets **1,500 sandbox minutes free each month**, renewing monthly. No credit card is needed — completing account verification is enough. It applies to all accounts, not just new ones, and your remaining balance is shown in the console.

* **Minutes are weighted by the size of the sandbox.** A sandbox up to 2 vCPU / 4 GB is fully covered; smaller configurations burn fewer minutes, and larger ones draw down proportionally more
* **Disk inside the free tier is free.**

See [Sandbox Free Tier](/agentic-studio/overview/getting-started/sandbox-free-tier.md).

**Pause and resume now take care of themselves.**

* **Paused sandboxes resume automatically** on their first request — no explicit resume call
* If more requests arrive than can be held while a sandbox wakes, the extra ones receive a **429** so your client can retry
* **Auto-pause is on**, so an idle sandbox pauses without you having to manage it
* **Set an idle window per sandbox**, so you decide how long it waits before pausing

See [Auto Pause and Resume](/agentic-studio/overview/sandbox-management/auto-pause-and-resume.md).

**New templates and everyday conveniences.**

* **File-based snapshots** capture just your workspace files, and can be taken and managed separately — including from neev-cli
* **Run Docker inside a sandbox** with an opt-in template, or use the new **Playwright + Chromium** template for browser automation
* **Labels** on sandboxes, with a label filter on the sandbox list
* **Resumable uploads** for large files
* **Reach a sandbox by name**, as well as by its ID
* **Expose and revoke preview ports** from the console. See [Exposing Ports for Preview URLs](/agentic-studio/overview/development-and-connectivity/exposing-ports.md)
* A **terminal tab** in the console for sandboxes and web agents

**Connect your own MCP client.** Sandboxes are now reachable through a single unified MCP endpoint, so an MCP-aware client such as Claude Code can create, run and manage sandboxes directly.

* Agents can **create sandboxes from any template** through the endpoint
* **New file and process tools**, and requests of **up to 1 MiB**
* Tools can be **discovered without an API key**, so a client can see what is available before you connect it

See [Sandbox MCP Server](/agentic-studio/overview/integrations-and-support/sandbox-mcp-server.md).

**Snapshots and restores are substantially more dependable.**

* **A restore never silently starts an empty sandbox.** If your saved state cannot be brought back, the sandbox fails with a clear error instead of coming up blank
* **Taking a snapshot no longer risks stopping your sandbox** or emptying its working files partway through capture
* **Partial captures are discarded** rather than offered later as a restore point that cannot work
* **Rollback works for small snapshots** — rolling back a sandbox that captured under 1 GiB of memory no longer fails
* **Running sandboxes are no longer removed** by the retention sweep that cleans up paused ones
* Sandboxes handed out from the **pre-warmed pool can now restore from a snapshot**, which previously failed for them
* **A resume that arrives while a pause is still finishing** now waits for it rather than racing it
* **Deleting a paused sandbox** now releases its quota straight away

See [Snapshots](/agentic-studio/overview/sandbox-management/snapshots.md).

**Quotas and visibility.**

* **Snapshot quotas** now apply: **100 per project** and **500 per organisation**
* **Quota errors name the scope they come from**, so you can tell a project ceiling from an organisation one
* The **Sandbox overview** shows **fleet usage across the project**, and sandbox state stays consistent between the list and the detail view

**Stateful code execution.** On the `sb-ubuntu-26-04-interpreter` template, `sandbox.code.run()` in the JS and Python SDKs (0.9.0) runs Python in a kernel that keeps its state between calls, with separate contexts when you need more than one.

### Agentic Studio — Agents (Beta, updated)

Agents catch up with sandboxes this release, and the most popular open-source agents now work without any provider setup.

* **Agents that work out of the box.** OpenClaw, Claude Code, Codex CLI, opencode and Hermes come preconfigured to use the NeevCloud Model API, with no provider to set up and OpenClaw's bundled plugins switched on. See the [Agents Catalogue](/agentic-studio/overview-1/catalogue.md)
* **Paused agents resume automatically** on their first request, just as sandboxes do
* **Fork a running agent** from the console to branch it into a copy and keep working
* **Roll an agent back to a snapshot** from neev-cli
* **Set an idle window per agent**, so you decide how long it waits before pausing
* **A clearer Agents console** — choose the provider when you create an agent, see usage and resources on the overview, inspect snapshot details, and follow a new **Pausing** status
* **MCP cookbooks for OpenClaw** show how to connect it to [Gmail](https://docs.ai.neevcloud.com/tutorials/openclaw-integrations/connect-gmail-mcp-to-openclaw-on-neevcloud), [Google Calendar](https://docs.ai.neevcloud.com/tutorials/openclaw-integrations/connect-google-calendar-mcp-to-openclaw-on-neevcloud), [Mattermost](https://docs.ai.neevcloud.com/tutorials/openclaw-integrations/connect-mattermost-mcp-to-openclaw-on-neevcloud) and [your sandboxes](https://docs.ai.neevcloud.com/tutorials/openclaw-integrations/connect-sandbox-mcp-to-openclaw-on-neevcloud)
* **In-place resize** and **live egress updates** apply to agents as well as sandboxes
* **Agent templates are refreshed** to the latest upstream versions of the tools they ship with
* **Pausing an agent that is already paused** no longer leaves it stuck

> Agents remain in Beta.

***

### AI Inference

* `/v1/responses` is available alongside `/chat/completions`, so a client already written against the Responses API — including Codex CLI — can point at NeevCloud without being rewritten
* **`/anthropic/v1/messages` support for GLM and MiniMax models**, which means you can point Claude Code directly at the Model API

See the [Model Routing API reference](https://docs.ai.neevcloud.com/api-reference/model-routing).

***

### Network Egress Controls

* **Update egress rules on a running sandbox or agent** and have them take effect immediately, instead of recreating the workload to change its network policy
* **Set ports and protocol per host** in egress rules, from the console
* **Route sandbox egress through your own machine or VPN**

See [Internet Access and Egress](/agentic-studio/overview/security-and-monitoring/internet-access.md).

***

### Developer Tools

* **Resize a sandbox or agent in place** from the JS and Python SDKs, without recreating it and losing state. See [Create, Pause, Delete Sandboxes](/agentic-studio/overview/sandbox-management/create-pause-delete-sandboxes.md)
* **Set lifecycle windows and create-time options** when you create a sandbox, so its lifetime is defined up front rather than managed by hand
* **ssh-agent forwarding** into sandboxes, so you can use your local keys without copying them in
* **Consistent error responses** across the platform, with a published list of error codes
* The **JS and Python SDKs now report a sandbox's last crash**
* **NeevCloud SDKs 0.9.0** (`neevai` and `@neevcloud/sdk`):
  * **Sandbox tools for your own agent** — `neevai.agent_tools` and `@neevcloud/sdk/agent-tools` give a model tools to run commands, work with files, expose ports, checkpoint and roll back, with adapters for the OpenAI Agents SDK, Claude Agent SDK, Vercel AI SDK and LangChain. See [Build Your Own Agent](/build-your-own-agent/overview.md)
  * **Snapshot and fork an agent** from the SDKs
  * **Labels** on create and as a list filter, and a wait for a snapshot to be ready
  * **Upgrading:** the audit trail now returns events in `data` instead of `records`, and `replicas` is removed in favour of `phase`. See the SDK changelogs for details

***

### Billing & Credits

* **Paused sandboxes are held for 14 days**, along with their snapshot
* **An alert 7 days before** your promotional or free credits run out
* **Promotional credits** and a **referral programme**, with a reward when someone you refer qualifies. See [Promotional Credit System](/getting-started/promotional-credit-system.md) and [Referral Program](/getting-started/referral-program.md)

***

### Account & Console

* **Business verification** is now available, alongside the individual flows, for both Indian and international customers. See [KYC Verification](/getting-started/kyc-verification.md)
* **Refreshed console pages**, including Organisation, Members, API Keys, Storage, Model API, Playground, Projects, Support and Profile
* **Support ticket statuses** in the console now match what our support team sees

***

## Known Issues

### VM Service

* **Boot time increases with GPU count.** A VM with more GPUs attached takes noticeably longer to come up. We are working on reducing this.

### Agentic Studio — Sandboxes

* **Older snapshots may not restore.** A snapshot taken on an earlier platform version can be incompatible with the current restore path. Where that happens, you are told before the restore is attempted; take a fresh snapshot to carry the state forward.
* **Writing a file larger than your allocated storage** does not fail immediately — you will see the error only after the sandbox stops for exceeding its limit. We are still working on catching this earlier.
* **Forking through the SDK** is still being stabilised. Forking through the CLI is unaffected.
* **The list panel does not re-select** when you switch projects, so you may need to pick a sandbox or agent by hand. A fix is planned for an upcoming release.

### Agentic Studio — Agents

* You may occasionally notice a short delay before the Agents view refreshes automatically.

### AI Inference

* When manually choosing the number of GPUs for a deployment, only certain counts (1, 2, 4, 8, and so on) are supported today. Choosing an unsupported number can cause the deployment to fail. We recommend using the automatic configuration option, which picks a supported count for you.
* The benchmarking view can currently overlap with the model details view. A cleaner layout is coming soon.

***

## Notes

August made sandboxes start in under a second and gave them snapshot, fork and rollback. Since then the work has gone into making that foundation usable without friction. The snapshot path was rebuilt so a restore is trustworthy — it brings back your state or says plainly that it cannot. Pause and resume now happen on their own, so an idle sandbox pauses without any lifecycle code on your side. And with 1,500 free minutes a month, you can try all of it before deciding whether it fits.

Agents now share most of that foundation. They resume on demand, fork and roll back, and the agents in the catalogue run against the Model API without any setup. The MCP endpoint is worth a look if you work in Claude Code or another MCP client: sandboxes become something you can drive from the tool you are already in, rather than a separate console.

Sandboxes and Agents remain in Beta.

As always, your feedback shapes what we build next. If you run into issues or have suggestions, reach out via a support ticket or contact your NeevCloud account team.

***

## Understanding Beta

Features marked as **(Beta)** are fully available to all users and ready to use, but are still being actively improved. Here is what to expect:

* **Core functionality works.** Beta features are not experimental - they are production-capable, but have not yet met all the stability and completeness thresholds required for General Availability (GA).
* **Frequent improvements.** You can expect updates to performance, reliability, and functionality in every release cycle.
* **Some gaps remain.** Certain views or flows may not be fully complete yet. Known gaps are listed in the Known Issues section above.
* **APIs may evolve.** Core endpoints are stable, but some parameters or response formats may be refined before GA.
* **No full SLA coverage.** Beta features are not covered by the same uptime commitments as GA features.
* **Your feedback matters.** Beta is a collaborative phase - the issues you raise and the workflows you share directly shape what we prioritise next. Please reach out.

Beta features graduate to GA when they meet our internal quality, performance, and completeness thresholds. GA graduation is always announced in the release notes.

***

*NeevCloud AI Platform · November 2026 ·* [*docs.ai.neevcloud.com*](https://docs.ai.neevcloud.com/)


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.ai.neevcloud.com/release-notes/nov-2026-release.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
