OpenClaw v2026.4.27 landed on April 29, 2026 at 5:12 p.m. America/Chicago time, which was April 29, 2026 at 22:12 UTC. The short version: Codex desktop control is easier to stand up, DeepInfra is now a first-class bundled provider, and a lot of the boring infrastructure work in this release is the kind that makes your next restart, channel reconnect, or update less likely to waste your evening.

This is not a tiny patch. It touches desktop control, provider choice, plugin metadata, startup behavior, Telegram and Slack reliability, Windows restarts, cron behavior, media tasks, and a pile of local-model edge cases. But the useful framing is simpler: you can do more out of the box, and the system is getting better at not fighting back.

Codex computer use is now much more real

The headline feature is the new Codex Computer Use setup flow. OpenClaw now ships status and install commands, marketplace discovery, optional auto-install, and fail-closed checks before a Codex-mode agent starts trying to drive your desktop.

That matters because desktop control is exactly the kind of feature that becomes dangerous when setup is half-manual and failure modes are vague. This release moves it closer to something you can verify before trusting.

/codex computer-use status
/codex computer-use install

If you have been curious about Codex-mode desktop control but did not want to spelunk through MCP wiring first, this is the build that makes another attempt reasonable.

DeepInfra is not just "another provider"

DeepInfra joins the bundled provider set with model discovery, image generation and editing, TTS, video, embeddings, and its own onboarding rules. That is a bigger deal than it sounds.

Adding a provider only helps if the rest of the product actually treats it like a native citizen. Here, OpenClaw did the work: catalog wiring, media flows, memory embeddings, and provider-owned setup policy all come along for the ride. If you like having one provider that can cover more than plain text, this release gives you a new serious option.

Startup work finally targets the stuff operators actually feel

A huge share of this release is startup and metadata cleanup. Plugin manifests keep absorbing catalog and activation metadata. Gateway startup reuses more cached manifest products. Channel startup stops waiting on the wrong work. Pricing refreshes, plugin scans, and model discovery all get a little less eager.

Most users will never read those bullets twice. They will notice the result the next time the gateway restarts faster, a channel comes back without drama, or a plugin-heavy install stops chewing CPU just to decide what it already knows.

There is also a lot less silent channel pain

Telegram got attention in several places: multi-account native approvals, outbound timeouts, 401 token failures, sticky network fallbacks, polling durability, webhook retries, and better typing/liveness behavior. Slack got fixes for socket mode stalls, media download hangs, and reset-trigger leakage. Discord picked up quieter reaction handling and sturdier interaction timing.

That is the kind of release work I like. It is not glamorous, but if your assistant lives in chat apps all day, it is the difference between "mostly works" and "I trust this enough to leave running."

What I would test right after updating

1. Run /codex computer-use status. If you plan to let Codex drive a desktop, verify the setup path before you need it under pressure.
2. Try one DeepInfra-backed workflow. Pick the thing you actually care about — text, image, TTS, or embeddings — and make sure the provider behaves the way you expect in your real config.
3. Restart the gateway once on purpose. This release spends a lot of effort on startup, plugin metadata reuse, and channel bring-up. It is worth checking that on your machine while you are watching.
4. Send one real message through your main chat channel. Especially if you rely on Telegram, Slack, or Discord, do one send and one reply cycle instead of trusting a green status badge.
5. If you run cron jobs, inspect one skipped or isolated job. Several cron fixes in this release are about failure classification, delivery routing, and timeout behavior. Better to verify that when you have context.
6. If you use local Ollama or other self-hosted providers, test one model that used to be flaky. There are enough local-model and fallback fixes here that a quick before-and-after check is worth your time.

Should you update now?

Yes if you care about Codex desktop control, want a new bundled provider with broad modality coverage, or run OpenClaw in a channel where reconnect weirdness has been costing you trust.

Also yes if you operate a customized install with plugins, cron jobs, local models, or multiple chat accounts. This release has a lot of maintenance-shaped work, and maintenance-shaped work is what keeps the next incident smaller.

This one will not impress you with a single giant headline after the first minute. It should impress you a few days later, when fewer things get weird.

---

Full changelog: View v2026.4.27 on GitHub

Generated release page: Read the structured release notes

- Fred