OpenClaw April 2026: Memory, Providers, and Operations

April's stable checkpoint, v2026.4.29, is about operating an agent after the novelty wears off. The release notes group the work around active-run steering, visible replies, a more inspectable memory layer, wider provider support, and sharper security checks.

Steer a run instead of starting over

Active runs now default to a steering queue. A follow-up can reach the next model boundary while the current work is still in progress, and the release documents the difference between steer, queue, and drop policies. Teams that need every reply to appear in the source chat can also require visible replies through the message settings.

That does not make an automation safe by itself. It does make the outcome easier to observe. Keep an audit trail for actions that change files, send messages, or spend money.

Memory adds provenance

The April notes describe a people-aware wiki, canonical aliases, relationship views, and evidence kinds. Active Memory can be limited to selected direct messages or conversations, while bounded partial recall can still return useful context when a hidden memory worker times out.

Those controls are a reminder to decide what should be remembered. Search quality is only half the problem; a private conversation that is recalled in the wrong channel is a privacy bug. Use the security guidance and test the allow and deny paths with non-sensitive data first.

Providers and plugins are more explicit

NVIDIA onboarding and catalog metadata arrive alongside safer model replay and streaming behavior. The release also adds a SQLite-backed plugin state store, so an installed plugin can keep a small keyed registry across restarts without sharing state with another plugin.

openclaw models
openclaw security audit

Use the model provider reference to separate provider setup from chat-channel setup. They are different trust and billing decisions.

April's operator checklist

• Choose a queue mode and write down which messages may interrupt a run.
• Limit Active Memory to conversations that need it.
• Review plugin state and provider credentials after an update.
• Run a test message through each channel that matters to you.

The full history remains in the official release index. Keep that link in your upgrade notes instead of relying on a static screenshot.