OpenClaw 4.7

Inference Hub, Memory Wiki & Enhanced Security

CLI
Inference Hub
WIKI
Memory Restored
50+
Security Fixes
NEW
Media Fallback

OpenClaw · Released April 2026 · v2026.4.7 · github.com/openclaw/openclaw

🚀 OpenClaw Infer — Unified Inference Hub

First-Class CLI for Provider-Backed Workflows

⚡ New openclaw infer command unifies model, media, web, and embedding tasks across all configured providers with automatic fallback and capability detection.

🤖 Model Inference

Run any configured model from CLI with streaming, reasoning control, and multi-turn conversations.

🎨 Media Generation

Generate images, music, and video with auto-fallback across providers, size/aspect remapping, and video-to-video support.

🌐 Web Search

Unified web search interface with configured provider fallback and consistent result formatting.

🔍 Embeddings

Generate embeddings for semantic search and RAG workflows with provider-agnostic interface.

Thanks @Takhoffman · Preserves intent during provider switches · Auto-remaps hints to closest supported options

📚 Memory Wiki Restored

Structured Knowledge Management with Claim Health & Freshness

🧠 Bundled memory-wiki stack returns with plugin, CLI, sync/query/apply tooling, structured claim/evidence fields, and freshness-weighted search.

📝 Structured Claims

Store knowledge as claims with evidence, sources, and confidence levels. Compiled digest retrieval for fast access.

🔍 Smart Search

Freshness-weighted search prioritizes recent knowledge. Contradiction clustering detects conflicting claims.

🏥 Health Monitoring

Claim-health linting identifies stale, unsupported, or contradictory knowledge. Staleness dashboards track decay.

🔄 Memory Host Integration

Seamless integration with memory-host for unified recall across wiki and daily notes.

Thanks @vincentkoc · CLI: openclaw memory wiki · Replaces deprecated memory-wiki plugin

🔌 Webhook Ingress Plugin

External Automation Meets TaskFlows

🌐 Bundled webhook plugin lets external systems create and drive bound TaskFlows through per-route shared-secret endpoints.

🔐 Secure Endpoints

Per-route shared secrets protect webhook endpoints. No public exposure without explicit configuration.

🔄 TaskFlow Binding

Webhooks can create, resume, and drive TaskFlow sessions with full context preservation.

🎯 Route Flexibility

Configure multiple webhook routes with different secrets, agents, and TaskFlow templates.

📊 Event Tracking

Built-in logging and monitoring for webhook invocations, failures, and TaskFlow state transitions.

Thanks @mbelinky · PR #61892 · Enables GitHub Actions, CI/CD, and external tool integration

💾 Session Compaction Checkpoints

Inspect & Restore Pre-Compaction State

⏪ Persisted compaction checkpoints + Sessions UI branch/restore actions let operators recover pre-compaction session state.

📸 Checkpoint Snapshots

Automatic snapshots before each compaction. Full conversation history preserved for recovery.

🌳 Branch & Restore

Sessions UI shows compaction points. Branch from any checkpoint or restore to previous state.

🔍 Inspection Tools

View full pre-compaction context, compare before/after, and audit what was summarized.

🔧 Pluggable Providers

New compaction provider registry lets plugins replace built-in summarization pipeline.

Thanks @scoootscooob, @DhruvBhatia0 · PR #62146, #56224 · Configure via agents.defaults.compaction.provider

🧠 Memory Dreaming (4.5)

Weighted Recall Promotion & Conceptual Tagging

💭 Experimental dreaming system promotes short-term memories to long-term storage with three cooperative phases: Light Sleep, Deep Sleep, and REM.
💤
Light Sleep
Quick scan of recent daily notes. Identifies candidates for promotion.
🌙
Deep Sleep
Consolidates related memories. Groups nearby daily-note lines into coherent chunks.
🌀
REM
Extracts lasting truths. Writes to MEMORY.md with evidence and confidence.

⚙️ Configurable Aging

Tune recencyHalfLifeDays and maxAgeDays to control recall decay and memory retention.

📖 Dream Diary

New Dreams UI surface shows dreaming trail in dreams.md. Preview REM staging with promote-explain.

Thanks @vignesh07, @davemorin · PR #60569, #60697 · CLI: /dreaming · Replaces competing dreaming modes

🎨 Media Generation (4.5)

Video, Music & Multi-Provider Support

🎬 Built-in video_generate and music_generate tools with bundled providers: Google Lyria, MiniMax, ComfyUI, xAI, Alibaba Wan, and Runway.

🎥 Video Generation

  • xAI grok-imagine-video
  • Alibaba Model Studio Wan
  • Runway Gen-3
  • ComfyUI workflows
  • Video-to-video support

🎵 Music Generation

  • Google Lyria (bundled)
  • MiniMax Music
  • ComfyUI workflows
  • Async task tracking
  • Follow-up delivery

🖼️ ComfyUI Plugin

  • Local & Comfy Cloud
  • Workflow-backed generation
  • Reference image upload
  • Live tests & output download

Auto-fallback across providers · Async completion with direct-send option · Unsupported hints ignored with warning

🌍 New Providers (4.5)

Expanded Model & Service Support

🤖 Chat Providers

  • Qwen — Alibaba's flagship models
  • Fireworks AI — Fast inference platform
  • StepFun — Chinese AI provider
  • Arcee AI — Trinity catalog + OpenRouter
  • Amazon Bedrock — Mantle support + auto-region

🔍 Search & Speech

  • Ollama Web Search — Local search integration
  • MiniMax Search — Chinese search provider
  • MiniMax TTS — Text-to-speech
  • Exa Search — Now visible in onboarding

☁️ Bedrock Enhancements

  • Inference-profile discovery
  • Automatic request-region injection
  • Claude, GPT-OSS, Qwen, Kimi, GLM support
  • Reduced manual setup

🔢 Bedrock Embeddings

  • Titan, Cohere, Nova, TwelveLabs
  • AWS credential-chain auto-detection
  • Provider-specific dimension controls

Thanks @wirjo, @arthurbr11, @eyjohn, @romgenie, @BruceMacD · Ollama vision capability auto-detection

🔐 Security Hardening (4.7)

50+ Security Fixes Across Attack Surfaces

🛡️ Major security release addressing SSRF, privilege escalation, token theft, and injection vulnerabilities across gateway, exec, browser, and channel plugins.

🚫 Exec Hardening

  • Block dangerous env overrides (Java, Rust, Cargo, Git, K8s, cloud creds)
  • Require owner auth for /allowlist changes
  • Fix Windows cmd.exe wrapper approval bypass
  • Block config.apply writes to exec approval paths

🌐 SSRF Protection

  • Drop request bodies on cross-origin 307/308 redirects
  • Treat main-frame redirects as navigations for private-network blocking
  • MS Teams file-consent upload URL validation
  • Enforce byte limits before base64 decode

🔑 Auth & Pairing

  • Invalidate stale WebSocket sessions on secret rotation
  • Require fresh pairing for nodes with new commands
  • Block cross-device token theft in pairing sessions
  • Fail closed on missing gateway pairing scopes

🔌 Plugin Security

  • Block persistent browser profile mutations via node.invoke
  • Verify ClawHub plugin SHA-256 before install
  • Block remote marketplace symlink escapes
  • Keep plugin routes on write-only scopes

Thanks @eleqtrizit, @pgondhi987, @obviyus, @ngutman · PRs #59119, #62002, #62291, #62357, #62355, #60489, #62658, #62350

🎯 Claude CLI Security (4.5)

Backdoor Session Hardening

🔒 Multiple security fixes prevent OpenClaw-launched Claude CLI sessions from being hijacked or redirected to attacker-controlled contexts.

🚫 Config Isolation

Clear inherited CLAUDE_CONFIG_DIR and CLAUDE_CODE_PLUGIN_* env vars so CLI runs can't be pointed at alternate config/plugin trees with different hooks or auth.

🔐 Provider Routing

Clear provider-routing and managed-auth env overrides. Mark runs as host-managed to prevent silent redirection to proxy, Bedrock, Vertex, or Foundry.

📁 Settings Sources

Force --setting-sources user even under custom backend args, blocking repo-local .claude project/local settings, hooks, and plugin discovery.

⚙️ Permission Mode

Treat malformed bare --permission-mode overrides as missing and fail safe to bypassPermissions, preventing flag consumption bugs.

Thanks @vincentkoc · Prevents backdoor session hijacking · Applies to all OpenClaw-launched Claude CLI runs

📱 Mobile & Platform (4.5)

iOS Exec Approvals, Matrix Support & UI Improvements

📲 iOS Exec Approvals

  • Generic APNs approval notifications
  • In-app exec approval modal
  • Works while iPhone locked/backgrounded
  • Apple Watch review & approval recovery
  • Notification cleanup on resolution

💬 Matrix Enhancements

  • Native exec approval prompts
  • Account-scoped approvers
  • Channel-or-DM delivery
  • Room-thread aware resolution
  • Auto-join setup with opt-in warnings

🌐 Control UI i18n

  • Simplified Chinese (zh-CN)
  • Traditional Chinese (zh-TW)
  • Brazilian Portuguese (pt-BR)
  • German, Spanish, Japanese, Korean
  • French, Turkish, Indonesian, Polish, Ukrainian

🎨 UI Improvements

  • Per-session thinking-level picker
  • Stop button visible during tool execution
  • ClawHub search in Skills panel
  • Light mode scrollbar fixes

Thanks @ngutman, @gumadeiras, @vincentkoc, @samzong, @chziyue · PR #60239, #58635, #60134

⚡ Performance & Stability

Prompt Caching, Provider Fixes & Channel Improvements

💾 Prompt Cache Optimization

  • More reusable prefixes across transport fallback
  • Deterministic MCP tool ordering
  • Normalized system-prompt fingerprints
  • Embedded image history support
  • openclaw status --verbose cache diagnostics

🤖 Provider Improvements

  • GPT-5 lower-verbosity defaults & visible progress
  • Claude thinking blocks preserved for 4.5+ models
  • Gemma 4 support with reasoning compatibility
  • Mistral Small 4 reasoning_effort support
  • Ollama multi-endpoint streaming fix

💬 Channel Fixes

  • Telegram: DM voice transcription restored
  • Discord: Reply tags, video splits, voice recovery
  • Slack: DM routing, thread stickiness
  • WhatsApp: Reconnect watchdog reset
  • Matrix: Secret storage recovery, DM sessions

🔧 Platform Stability

  • Gateway PID recycling detection (Windows/macOS)
  • macOS LaunchAgent recovery on restart
  • Windows scheduled task reinstall preservation
  • Docker/Podman auto-bind to 0.0.0.0
  • Node 22 undici HTTP/2 compatibility fix

Thanks @bcherny, @vincentkoc, @neeravmakwana, @sfuminya, @afurm, @MonkeyLeeT, @gumadeiras, @zozo123

🎓 Key Takeaways

OpenClaw 4.7 — Production-ready inference hub, restored memory wiki, and comprehensive security hardening.
OpenClaw 4.5 — Media generation (video/music), dreaming system, 12 UI languages, and mobile exec approvals.

🚀 For Developers

  • Use openclaw infer for unified inference workflows
  • Enable memory wiki for structured knowledge management
  • Configure webhook endpoints for CI/CD integration
  • Review security updates and update exec policies

🎨 For Creators

  • Generate videos with xAI, Runway, or Alibaba Wan
  • Create music with Google Lyria or MiniMax
  • Use ComfyUI workflows for custom media pipelines
  • Enable dreaming for automatic memory consolidation

📦 Upgrade: npm install -g openclaw@latest

Release Notes: v2026.4.7 · v2026.4.5
Twitter: @openclaw announcement