October 15, 2026

Policy changes

No charges while a host has a maintenance window open

When a host opens a maintenance window on a machine — scheduled or emergency — Pods on that machine (including dedicated serverless replicas) and its volumes are not billed until the window ends.

• Your Pod keeps running. Meshive doesn’t stop or pause it; it is simply free while the window is open.
• Billing resumes the moment the window ends, including when the host ends it early.
• What is still billed: tasks and stored assets are billed as usual. If the machine is a storage node, only the volumes stored on it are exempt — Pods on other machines that mount them are still billed.
• In the console, an affected Pod or volume shows a Host maintenance label and $0.000/hr while the window is open. The card stays usable, so you can keep connecting to a running Pod.

Billing and earnings during a window →

Major features

Python SDK, CLI and MCP server 1.2.0

1.2.0 is now on PyPI and Docker Hub, and the hosted MCP server runs it. It edits Pods and storage like the console, opens storage over SSH, starts Pods with a model set, and includes the features first announced as 1.1.0. See the October 14 notes for the full list. Upgrade with pip install -U meshive, or use the meshive/meshive-mcp:1.2.0 image if you run the MCP server yourself.

What changes today for every SDK, CLI and MCP version, because the server now enforces it:

• vCPU and RAM must be sizes the console offers, a port the template already opens can’t be added again, and environment variables the template requires need a value.
• CPU Pods created without a vCPU count get 8 vCPU and 32 GiB of RAM (or the nearest sizes the line offers) instead of 2 vCPU and 4 GiB. Pass a size if you want a smaller Pod.
• Storage created on its own is network storage of at least 10 GiB.
• Secret environment variables in Pod listings are masked as *** for read-only API keys. Read & write keys still see the login values shown in Connect; other secrets are masked for them too.
• When you attach only some files of an input asset, every path you list must match at least one file. A path that matches nothing is now refused instead of being skipped quietly, and file and folder names match however their non-Latin characters are encoded.

Python SDK → · CLI → · MCP server →

Improvements

• Faster asset downloads into Pods. Assets fetched when a Pod starts now download up to about twice as fast. This applies to Pods created, edited or moved from today.
• Downloads pick up where they stopped. If a Pod restarts while its assets are downloading, large files continue from where they stopped instead of starting over, and the finished files of a multi-file asset aren’t fetched again. A download that was cut short is detected and retried, and leftover partial files no longer take up space in your volume.
• Image download progress shows up sooner when a new Pod pulls its container image, and updates more steadily.
• Idle tabs stay quiet. The console stops refreshing while its tab is hidden, and in the client console after 30 minutes without activity. It catches up right away when you come back.

Fixes

• Renaming a volume from Edit works again. The new name is saved — a name another volume in the workspace already uses is refused — and the window opens with the volume’s current settings, so saving no longer brings back an old name or usage-warning setting.
• Usage warnings turned on when creating storage now take effect. The setting was being dropped, so no warning email was sent. Storage created before today keeps the old setting; turn the warning on again in Edit.

For hosts

• No earnings from a machine while its maintenance window is open, and no commission is taken. End the window from the Maintenance tab when you’re done to start earning again right away. The uptime rules for maintenance windows are unchanged. Billing and earnings during a window →