OpenClaw Updates: April 2026

OpenClaw shipped four releases in April 2026. This post covers the highlights from v2026.4.5 through v2026.4.9: new media generation tools, memory and dreaming improvements, security fixes, and Control UI refinements. If you run OpenClaw in production or are planning to deploy it, these are the changes worth knowing about.

Media generation: video and music

v2026.4.5 introduced built-in video and music generation tools. Agents can now create videos and audio tracks through configured providers.

• Video: New video_generate tool with support for xAI (grok-imagine-video), Alibaba Model Studio Wan, and Runway.
• Music: New music_generate tool with Google Lyria, MiniMax, and workflow-backed Comfy support.
• Both tools include async task tracking and follow-up delivery when generation completes.

These tools follow the same pattern as the existing image generation tool: the agent requests media, the gateway dispatches to the configured provider, and the result is returned when ready. No external service setup is required beyond your existing API keys.

Memory and dreaming improvements

The experimental dreaming feature received significant updates across v2026.4.5 and v2026.4.9.

v2026.4.5: dreaming foundation

• Weighted short-term recall promotion
• New /dreaming command and Dreams UI
• Multilingual conceptual tagging
• Three cooperative phases (light, deep, REM) instead of competing modes
• Configurable aging controls: recencyHalfLifeDays, maxAgeDays
• REM preview tooling and deep promotion replay safety

v2026.4.9: grounded backfill and diary

• Grounded REM backfill lane with historical rem-harness support
• Diary commit/reset flows
• Cleaner durable-fact extraction with live short-term promotion integration
• Old daily notes can replay into Dreams without requiring a second memory stack
• New Control UI diary view with timeline navigation and backfill controls

The dreaming system is still experimental, but these updates make it more reliable for operators who want durable memory promotion running in the background. If you tried dreaming in earlier versions and found it unstable, the v2026.4.9 improvements are worth testing again.

Security fixes

v2026.4.9 shipped several security fixes that matter for production deployments:

• Browser SSRF protection: Blocked-destination safety checks now re-run after interaction-driven navigations (click, evaluate, hooks). Browser interactions can no longer bypass SSRF quarantine when landing on forbidden URLs.
• Dotenv security: Runtime-control env vars and browser-control overrides are now blocked from untrusted workspace .env files.
• Node exec events: Remote node exec events are marked as untrusted and sanitized before enqueueing, preventing remote node output from injecting trusted system content.
• Plugin auth collision prevention: Untrusted workspace plugins can no longer collide with bundled provider auth-choice IDs during onboarding.

If you run OpenClaw with browser automation or remote nodes, upgrading to v2026.4.9 is recommended.

Provider updates

New bundled providers

• Amazon Bedrock: Full support including inference profiles and automatic request-region injection for Claude, GPT-OSS, Qwen, Kimi, and GLM models.
• ComfyUI: Workflow media plugin for local ComfyUI and Comfy Cloud with image, video, and music generation support.
• MiniMax: Image generation for image-01 model with aspect ratio control, plus TTS integration.
• Qwen, Fireworks AI, StepFun: New bundled providers for chat and completion workflows.

Prompt caching improvements

Multiple changes across v2026.4.5 improve cache hit rates and reduce costs:

• Reusable prompt prefixes across transport fallback
• Deterministic MCP tool ordering
• Normalized system-prompt fingerprints
• Cache diagnostics via openclaw status --verbose
• Removal of duplicate in-band tool inventories from agent system prompts

Control UI and platform updates

• Multilingual UI: Control UI now supports Simplified Chinese, Traditional Chinese, Brazilian Portuguese, German, Spanish, Japanese, Korean, French, Turkish, Indonesian, Polish, and Ukrainian.
• ClawHub integration: Skills panel now includes ClawHub search, detail, and install flows.
• iOS exec approvals: APNs notifications with in-app approval modals.
• Matrix exec approvals: Native Matrix approval prompts with account-scoped approvers.
• Android pairing: Improved reliability with stale auth clearing and auto-retry pausing when backgrounded.

Channel fixes

v2026.4.8 and v2026.4.9 include important fixes for bundled channels:

• Telegram: Fixed setup contract loading for npm installs, voice-note transcription in DMs, and native command menu trimming.
• Slack: Fixed proxy settings for Socket Mode WebSocket connections, bearer auth preservation across redirects, and resolved token handling for file downloads.
• Matrix: Improved sync readiness handling, contained background handler failures, and added DM session scoping.
• WhatsApp: Fixed echo loops in self-chat mode and restored watchdog timeout handling.

How to update

If you installed via npm:

npm install -g openclaw@latest

If you use the Docker image:

docker pull openclaw/openclaw:latest

After updating, verify the install:

openclaw version openclaw doctor

What to watch

The dreaming feature is still marked experimental. If you enable it, monitor your memory usage and check the Dreams UI regularly. The backfill functionality in v2026.4.9 makes it easier to recover from failed promotions, but it is not yet a replacement for manual MEMORY.md maintenance.

Media generation tools are stable but provider-dependent. If you use xAI, Runway, or Alibaba for video generation, test with small prompts first to verify your API keys and rate limits.