OpenClaw v2026.4.9 launches 'Dreaming' feature for AI agent memory

OpenClaw shipped version 2026.4.9 on April 9, introducing “Dreaming” — a memory feature that lets agents process historical data through REM backfilling and replay user notes to form persistent memories.

What shipped

The v2026.4.9 release includes three major additions:

• REM backfill: Agents can now process historical conversation data during idle periods, building richer context without consuming active compute
• Diary timeline UI: A new journal-style interface lets users see what their agent “remembers” and how memories form over time
• Security hardening: Patches for SSRF and node execution injection vulnerabilities identified in the ongoing security audit cycle

Why it matters

Memory has been OpenClaw’s weak spot compared to competitors like Hermes Agent, which ships with a native learning loop. The Dreaming feature closes that gap by giving OpenClaw agents the ability to build context autonomously — a first for the platform’s 120K+ star ecosystem.

This also signals the Foundation’s post-Steinberger direction: rather than chasing new integrations (OpenClaw already has 50+), they’re investing in agent intelligence and autonomy.

What to watch

The v4.0 milestone release is expected mid-2026 with a redesigned plugin SDK, multi-agent orchestration, and a new web dashboard. The Dreaming feature will likely be central to how OpenClaw positions itself against Hermes Agent’s self-improving capabilities.