HeadsUpAI

OpenClaw v2026.4.29 Makes AI Agents Native to Group Conversations

OpenClaw, an open-source self-hosted AI assistant, released version 2026.4.29 to improve group chat orchestration. The update introduces active-run steering, a queueing logic that allows new messages to influence an agent's current task at the next model boundary (the point where a model finishes one step).
Steering fallback debounce
500ms
New locales
Persian, Dutch, Vietnamese, Italian, Arabic, Thai
Memory recall timeout
150s maximum budget
NVIDIA integration
API-key onboarding and static catalogs
Security
OpenGrep scanning and GHSA triage policy

This release addresses the friction of multi-user environments where rapid-fire context often leads to redundant loops. By evolving the Memory Wiki knowledge layer into a people-aware system with relationship graphs, the platform moves closer to a persistent digital colleague. It also expands infrastructure flexibility by adding native support for NVIDIA model catalogs.

You can deploy the update via the openclaw update command to access the new SQLite-backed plugin state store. The release also includes parity for Claude Opus 4.7 thinking profiles on Amazon Bedrock and opt-in follow-up commitments for automated reminders. The software remains open-source for local or server-based installation.

View the full update on github.com
OpenClaw🦞
OpenClaw🦞
@openclaw
X

OpenClaw 2026.4.29 🦞 💬 Group chats feel much better now 📌 Follow-up commitments from context 🔐 Safer exec, pairing, and owner controls 🟩 NVIDIA provider + model catalogs ⚡ Faster startup + plugin/channel fixes Group chat finally feels agent-native. https://t.co/C6GXdLeakz

120retweets1.1klikes
View on X

Still wondering? A few quick answers below.

Active-run steering is a new queueing mechanism that allows users to influence an agent's current task while it is still running. Instead of spawning duplicate work or waiting for a loop to finish, the agent processes follow-up messages at the next model boundary, which is the point where the AI completes a discrete reasoning step.

The Memory Wiki has evolved into a people-aware knowledge system that uses relationship graphs and canonical aliases to organize information. It includes provenance reports that track the origin of specific claims and evidence-kind drilldowns. This allows the AI to maintain a structured understanding of different individuals and their connections across multiple conversations.

Yes, the v2026.4.29 update adds a dedicated NVIDIA provider with API-key onboarding and static model catalogs. Users can now select NVIDIA-hosted models directly through the platform's interface. This integration includes faster manifest-backed model paths and literal model-reference picker support, allowing the assistant to use specialized hardware or hosted NVIDIA NIM models for inference.

Follow-up commitments are an opt-in feature where the AI assistant automatically infers obligations or reminders from conversation context. These commitments are tracked using hidden batched extraction and delivered later via heartbeat intervals. Users can manage these tasks through a command-line interface and configure limits on the maximum number of automated check-ins allowed per day.

OpenClaw is an open-source, self-hosted AI assistant that can be installed on various operating systems. To update to the latest version, users can run the openclaw update command in their terminal. This process synchronizes plugins and refreshes the runtime environment, including the new SQLite-backed state store that provides restart-safe keyed registries for plugin data.

Share this update