I have been trying to find a permanent AI sidekick this year and know the headache that comes when you think about migrating. Export this, reformat that, lose half your setup in transit. So when an update makes switching easier instead of harder, it gets my attention.

OpenBot just shipped version 0.25.2, and the headline feature is aimed squarely at Grok Bot users: every member of a joined server can now import agents straight from Grok Bot. Not just the server owner. Everyone.

The mechanics are simple. On a joined server, open Server settings, hit Import, and hand it your Grok Bot export as a zip file under 100 MB. If the server already has a skill the export needs, the imported agent just uses it, and the import tells you so.

That matters because the hardest part of leaving any assistant was never the decision. It is the rebuild. Your agents, your routines, your little custom setups. An importer that preserves them removes the excuse to stay put.

It also shows where OpenBot thinks it wins. The project maintains a page comparing itself against Grok Bot, Muse, and the rest of the pack. This release reads like that page turned into a feature.

The Grok Bot import is the crowd-pleaser, but the GitHub changes might matter more day to day. You can now connect GitHub under Server settings, in the Connectors section, and every agent on that machine gets GitHub tools. No personal access token to create, copy, and lose.

In repos where you can push, anything an agent does with those tools shows up as openbotgit[bot] rather than under your name. That separation is the kind of detail you only appreciate after an agent has once committed something embarrassing as you.

Under the hood it runs on short-lived tokens, and OpenBot keeps none of them lying around. The gh and git commands on the machine still act as you. Sensible split.

Server owners get quality of life fixes too. Owners and admins can update joined hosts straight from Server settings, watch the download percentage, and restart when agents are idle. Nobody has to SSH anywhere.

Hosted servers keep growing up as well: real plans with billing, storage, and member limits, automatic sleep when idle, and automatic reconnects when needed. The pitch is a server you can forget about, in the good way.

Smaller, but I like it: agents that are waiting on you now gather in a "Needs you" group at the top of the sidebar, each row saying what it wants. An answer, a review, or a takeover. The red badge of shame, organized.

There is also a batch of first-run polish. New servers ask you to connect a provider before creating your first agent, and you can sign a headless server into Claude or Grok from your own device with a code. Fewer dead ends for newcomers.

On the fixes side, Grok chats survive account switches and key changes now. Previously every message in an old chat died with a cryptic encrypted content error. OpenBot now starts a fresh Grok session behind the scenes and keeps your history.

And when an OpenCode model request fails, the error finally says what happened: a rate limit, a billing problem, a provider outage, or no network, plus whether waiting helps. No more "Internal error" archaeology.

None of this is glamorous. Imports, connectors, error messages. But assistant apps live or die on exactly this layer, the unglamorous plumbing between you and the model.

OpenBot's whole bet is that the assistant should be a shared workspace for agents rather than a single chatbot. We covered the 0.24.0 release, which let agents build and manage each other, just days ago. Releases like this one are that idea compounding.

The full list is longer than the highlights here, and as usual the changelog tells you what to do after upgrading.

If you have been meaning to try life beyond Grok Bot, Muse, or Dots, this is a convenient week. The importer is ready, and your agents can come with you.