OpenClaw v2026.4.29 Released: Memory Upgraded to 'Character Wiki', Agent Learns to Proactively Track Commitments

According to monitoring by Dongcha Beating, the open-source personal AI assistant OpenClaw (GitHub 367K stars) has released v2026.4.29, marking its second update in two days. The most significant changes for users are in the memory system and messaging system. Memory has evolved from simple retrieval to a character perception wiki: agents can automatically build character cards and track interpersonal relationship graphs, with each memory including source tracing and evidence type annotations. Active Memory now includes filtering by conversation ID, allowing operators to enable memory recall only for specified conversations; sub-agents no longer discard retrieved content after timing out but instead return partial recall summaries. The messaging system has introduced a ‘commitment’ feature. Agents implicitly extract follow-up commitments during conversations (such as ‘remind you tomorrow’ or ‘check progress next week’), delivering them on a schedule via a heartbeat mechanism, with independent configuration limits for agents and channels. The queuing strategy for user messages has also changed: the default has switched from processing messages one by one to a steer mode, which reads all pending messages at once during model inference, reducing interruptions. Other changes include NVIDIA joining as a built-in model provider, allowing calls with an API Key; a new SQLite plugin for persistent state storage, ensuring no data loss upon restart; integration of OpenGrep rule scanning for security, automatically uploading SARIF to GitHub Code Scanning; plugin loading time on Windows reduced from about 39 seconds to approximately 2 seconds; and fixes for over ten Block Kit overflow issues in the Slack channel.