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

According to Dynamic Insight Beating monitoring, the open-source personal AI assistant OpenClaw (GitHub 367K stars) has released v2026.4.29, its second update within two days. The most significant changes for users are in the memory system and message system.

The memory system has evolved from simple retrieval to character-aware wiki: the agent can automatically build character cards, track interpersonal relationship graphs, and each memory includes source tracing and evidence type annotations. The Active Memory feature now includes filtering by conversation ID, allowing operators to selectively enable memory recall for specific conversations only; sub-agents no longer discard retrieved content when timing out, but instead provide a partial recall summary.

The message system has introduced a "Commitment" feature. The agent implicitly extracts follow-up commitments during conversations (such as "I will remind you tomorrow" or "Check progress next week"), delivers them periodically through a heartbeat mechanism, and supports independent configuration limits by agent and channel. The message queuing strategy for user messages has also been updated: it has switched from default per-message queuing to a steer mode, which reads all pending messages at once at the model's inference boundary to reduce interruptions.

Other changes include: NVIDIA joining as an in-house model provider, accessible with a configured API Key; plugins now feature SQLite persistent state storage to prevent data loss upon restarts; integrated OpenGrep rule scanning for security, automatically uploading SARIF to GitHub Code Scanning; plugin loading time reduced from around 39 seconds to about 2 seconds on Windows; over a dozen Block Kit limit overflow issues fixed for the Slack channel.