OpenClaw v2026.4.5 is the biggest release of April 2026 so far. It turns agents into multimedia creators, expands provider options significantly, localizes the entire control UI, and introduces an experimental memory system inspired by how humans consolidate memories during sleep.
Two new tools ship out of the box: video_generate and music_generate. Any agent can now create video clips and music tracks as part of its workflow.
A content agent could write a blog post, generate an accompanying video summary, and create background music for it — all in a single conversation. A marketing agent could produce short-form video ads from product descriptions. The tools are provider-agnostic, so you connect whatever video or music API you prefer.
This is a major shift from text-only agent capabilities. OpenClaw agents are no longer just chatbots that can run code — they are multimedia production pipelines.
v2026.4.5 adds first-class support for four new LLM providers:
With these additions, OpenClaw now supports over 20 LLM providers out of the box. The multi-provider architecture means you can mix and match — use Claude for reasoning, Qwen for Chinese content, and Fireworks for high-volume batch processing, all within the same agent team.
The entire OpenClaw control interface is now available in 12 languages: Simplified Chinese, Traditional Chinese, Brazilian Portuguese, German, Spanish, Japanese, Korean, French, Turkish, Indonesian, Polish, and Ukrainian.
This is not just button labels — settings panels, error messages, system notifications, and documentation links are all translated. OpenClaw has always been popular internationally, but this is the first release where non-English speakers get a fully native experience.
Language detection is automatic based on system locale, with a manual override in settings. Community contributions for additional languages are welcome through the standard translation workflow.
The Matrix messaging integration now supports native exec approval prompts. When an agent needs to run a shell command, the approval request appears as a native Matrix message with approve/deny buttons — no need to switch to the OpenClaw UI.
This brings Matrix to feature parity with Telegram and Discord for agent management. Teams using Matrix as their primary communication tool can now fully control agent permissions from within their chat client.
OpenClaw now supports Apple Push Notification service (APNs) for iOS devices. When your agent completes a task, encounters an error, or needs approval, you get a native iOS notification.
Combined with the Android notifications from earlier releases, this means OpenClaw agents can reach you on any mobile device. No more checking a dashboard — your agents notify you directly.
The most fascinating feature in v2026.4.5 is memory dreaming — an experimental system that consolidates agent memory during idle periods, inspired by human sleep cycles.
It runs in three phases:
The result is an agent that gets smarter over time without explicit retraining. After a dreaming cycle, agents show improved recall accuracy and can make connections they would have missed otherwise.
This is marked experimental for a reason — the REM phase can occasionally create false associations. But early testing shows promising results, especially for research and analysis agents that accumulate large knowledge bases.
Prompt caching has been flaky in recent releases, especially with multi-turn conversations. v2026.4.5 stabilizes the caching layer with better invalidation logic and consistent cache hit rates across all providers that support it.
For high-volume deployments, this translates directly to lower API costs and faster response times. Prompt caching can reduce token usage by 50-80% for repetitive agent interactions.
Standard update process — pull the latest image or update your installation. Memory dreaming is disabled by default; enable it in experimental settings if you want to try it. The new providers and multimedia tools are available immediately after update.
Full changelog on the GitHub release page.
The video_generate tool is provider-agnostic. It works with any video generation API you configure — Runway, Pika, StepFun, or custom endpoints. OpenClaw handles the prompt formatting and result delivery; you choose the backend.
Memory dreaming is an experimental feature that consolidates agent memory during idle periods. It runs in three phases: light (recent memory indexing), deep (cross-referencing old and new memories), and REM (creative connections with weighted recall). Think of it as offline processing that makes your agent smarter over time.
v2026.4.5 adds Simplified Chinese, Traditional Chinese, Brazilian Portuguese, German, Spanish, Japanese, Korean, French, Turkish, Indonesian, Polish, and Ukrainian. The control UI, settings panels, and system messages are all translated.
No. Qwen, Fireworks AI, StepFun, and MiniMax are bundled as provider plugins. Just add your API key and select the provider. Model routing and prompt formatting are handled automatically.
Skip the setup. Pick a template and deploy in 60 seconds.
Pick a role. Your AI employee starts working in 60 seconds. WhatsApp, Telegram, Slack & Discord. No setup required.
Get Your AI Employee