OpenClaw shipped a task board. That sounds minor. It is not.
Version 2026.4.1 includes /tasks, a chat-native background task board for the current session. Agents can now track task details, show fallback counts when no linked tasks are visible, and maintain a persistent record of what they were working on within a session. Before this, OpenClaw agents could execute tasks but had no built-in way to track which ones were running, completed, or stalled. The task board closes that gap.
For builders, that means agents can now handle workflows that require twelve tool calls across four different services and report their own progress without you asking. A data pipeline that runs for three days can surface failures at step seven without you digging through logs. An onboarding workflow that sets up accounts, permissions, and access across five systems simultaneously becomes observable in real time. The agent shows you which steps succeeded, which failed, and which are still running, inside the same chat thread where the work is happening.
The task board PR was contributed by a community member, not the OpenClaw core team. That is worth noting: it means the feature was wanted badly enough that someone outside the team built it. In open-source projects, community contributions filling gaps the core team did not prioritize first is a healthy signal.
Two other changes in 2026.4.1 matter for production deployments. The agents failover params add a new knob, auth.cooldowns.rateLimitedProfileRotations, that controls how many times an agent retries the same provider profile before attempting a cross-provider fallback when it hits a rate limit. Before this, the retry behavior was not configurable — a stalled stream would hang until the broader run timeout fired. For agents running critical workflows, this is the difference between graceful degradation and a silent failure that goes unnoticed until someone checks.
The cron tools allowlist lets you restrict which tools a given cron job can call on a per-job basis. This is a direct response to CVE-2026-25253, the critical OpenClaw vulnerability from earlier this year. Before this change, a compromised or misconfigured cron job had access to the full toolset. Now you can say a cron job only has access to the web search tool and nothing else, which limits the blast radius if something goes wrong.
AWS Bedrock Guardrails support in the bundled provider, a configurable webchat history truncation parameter, and consistent model override handling across manual and automatic context compaction paths round out the notable additions.
The full changelog is on GitHub. The task board is the headline. It is the first time a persistent working-memory equivalent has existed inside OpenClaw's chat interface.
OpenClaw 2026.4.1 release notes are at github.com/openclaw/openclaw/releases/tag/v2026.4.1.
† Add † footnote: "Source-reported; not independently verified."
† Add † footnote: "Source-reported; not independently verified."
