Release Overview
OpenClaw 2026.4.29 is a landmark update for contextual intelligence and infrastructure reliability. By turning memory into a People-Aware Wiki and adding native support for NVIDIA's model catalog, this release significantly raises the ceiling for what autonomous agents can achieve in social and high-performance environments.
Memory Evolution: The People Wiki
Memory is no longer just a flat list of facts. In 2026.4.29, OpenClaw introduces a structured, person-centric memory system.
- Person Cards: Agents now maintain detailed cards for people they interact with, including canonical aliases and relationship graphs.
- Provenance & Evidence: Every claim in the memory wiki tracks its source, allowing agents to explain why they believe something.
- Active Filtering: New
allowedChatIdsanddeniedChatIdsfilters give you granular control over what gets committed to long-term memory.
NVIDIA & High-Performance AI
Native NVIDIA support brings massive scale to OpenClaw:
- Direct Onboarding: Integrated API-key setup and model discovery for the full NVIDIA catalog.
- Bedrock Opus 4.7: Full parity with the latest 'thinking' models on AWS Bedrock.
- Faster Manifests: Optimized model and auth paths reduce gateway startup time and model switching overhead.
Messaging & Automation Hardening
Communication reliability is the backbone of OpenClaw. This release fixes critical edge cases across every major channel.
- Steering by Default: Active-run steering is now the default for messaging and automation, ensuring agents remain responsive during long tasks.
- Visible-Reply Enforcement: New
messages.visibleRepliessetting ensures agents always provide feedback in source chats. - Channel Resilience: Major fixes for Slack Block Kit limits, Telegram webhooks, Discord rate-limiting, and WhatsApp liveness.
Security & Operations
OpenGrep Scanning
OpenClaw now uses OpenGrep for internal security scanning. This includes a precise rulepack that validates first-party code and ensures that changes to the core system meet the highest safety standards.
Safer Tool Handling
Tool profiles have been hardened. Configured tool sections (like exec or fs) no longer implicitly widen restrictive profiles. This "fail-secure" approach ensures that minimal and messaging profiles stay restricted unless explicitly granted more access.
Upgrade Guide
Who should upgrade?
- Operators managing agents in complex social/group environments
- Users utilizing NVIDIA or Bedrock Opus 4.7 models
- Teams requiring visible-reply guarantees and task steering
- Anyone prioritizing security with OpenGrep-validated code
How to Upgrade
# Update to 2026.4.29
openclaw update
# Verify memory wiki status
openclaw memory statusFor the complete list of 80+ fixes and technical changes, visit the official release page on GitHub.
FAQ
What is the new People-Aware Memory Wiki?
OpenClaw memory has evolved into a structured wiki that understands people. It tracks relationships, person cards, and claims with full provenance, allowing agents to better understand the social context of their conversations.
How do I use the new NVIDIA provider?
The NVIDIA provider is now built-in with an easy onboarding process. You can select models directly from the NVIDIA catalog using their provider prefix after adding your API key.
What is Active Memory Filtering?
You can now control which chats your agent remembers. Active Memory filters allow you to opt-in or opt-out of specific direct messages, groups, or channels, ensuring privacy and context relevance.
What is visible-reply enforcement?
A new global setting (messages.visibleReplies) allows operators to require that all agent output is sent as a visible message. This prevents agents from completing turns silently, especially in complex group chat scenarios.
Need help from people who already use this stuff?
Master Agent Memory
The new People-Aware Memory Wiki is a game changer for social agents. Join the community to see how others are building relationship graphs and leveraging NVIDIA models.