OpenClaw 2026.4.1: Smarter Task Management, Stronger Cross-Platform Stability, and a More Reliable Agent Experience

We are pleased to announce the release of OpenClaw 2026.4.1. This version focuses on refining the everyday user experience, strengthening core functionality, and delivering a more stable and predictable AI assistant—whether you are an individual developer or part of a larger team.

This update spans several key areas: task management, cross-platform integration, command execution security, and system stability. It also addresses a number of issues reported by the community. Below, we walk through the most important changes and what they mean for your workflow.

A Clearer View of Background Tasks

If you often rely on OpenClaw to handle complex background operations, the new /tasks command will feel like a natural extension of your workflow. Typing /tasks in any chat session now opens a session-native task board. This interface gives you a real-time overview of all tasks associated with the current session, including:

  • Active tasks and their status
  • Recently completed tasks with relevant details
  • Agent-local fallback counts when no linked tasks are visible

This shift makes background task management transparent and accessible. Instead of guessing what the agent is doing or relying on complex status checks, you get a clean, readable summary directly in the conversation.

Deeper Integration Across Platforms

OpenClaw continues to expand its ecosystem, and this release brings meaningful improvements to several platforms.

Feishu Drive Collaboration

For teams using Feishu (Lark), the new Drive comment-event flow represents a significant step forward. OpenClaw can now:

  • Parse comment threads to understand the full context of a discussion
  • Reply directly within the thread, keeping conversations coherent
  • Use dedicated feishu_drive comment actions to participate in document collaboration workflows

This turns the agent into a more natural participant in document reviews and collaborative writing, where it can respond to feedback, ask clarifying questions, or suggest edits without leaving the document environment.

macOS Voice Wake

macOS users now have a hands-free way to interact with OpenClaw. The new Voice Wake option allows you to trigger Talk Mode using a voice command. This is especially useful when your hands are occupied or when you want a more seamless, voice-first interaction with the agent.

WhatsApp and Telegram Refinements

On WhatsApp, OpenClaw now passes the inbound message timestamp to the model context. This seemingly small change allows the AI to understand when a message was sent, enabling more contextually appropriate responses—for example, acknowledging a late-night message appropriately.

On Telegram, we added configurable errorPolicy and errorCooldownMs controls. These give you better control over how the agent handles repeated delivery errors. Instead of flooding a chat with identical error messages, the agent can suppress repeated failures while still reporting new or distinct issues. This keeps group chats and channels clean without hiding genuine problems.

Expanded Model Support

The model catalog continues to grow. This release adds glm-5.1 and glm-5v-turbo to the bundled Z.AI provider catalog. Having more model options means you can better match the model to the task at hand—whether you need pure text generation or multimodal understanding.

Smarter, Safer Command Execution

Command execution approvals have been a focus area for several releases. With 2026.4.1, we revisited the approval logic to make it more predictable and secure.

What Changed in Execution Approvals?

We addressed a number of edge cases that could lead to confusing behavior:

  • allow-always is now truly persistent. Previously, marking a command as “always allow” sometimes behaved like a one-time approval. This has been corrected. Once you approve a command permanently, that trust is durable.
  • Exact-command trust is reused. For shell-wrapper paths where a persistent allowlist entry cannot be safely stored, the system now reuses exact-command trust records. This reduces the number of redundant approval prompts.
  • Static allowlist entries no longer bypass ask:"always". If a rule is configured to always ask for approval, static allowlist entries will not silently override it.
  • Windows execution is more robust. When Windows cannot build an allowlist execution plan, the system now requires explicit approval rather than failing silently or dead-ending the remote execution.

Better Visibility with openclaw doctor

The openclaw doctor command now includes a warning when tools.exec is configured more broadly than the rules in ~/.openclaw/exec-approvals.json. This gives you a clear signal when your host-level policies might be less restrictive than your per-command approvals, helping you maintain a consistent security posture.

Smarter Failover and Rate Limit Handling

When a model provider returns rate-limit errors, OpenClaw now handles them more intelligently. The new auth.cooldowns.rateLimitedProfileRotations setting allows the system to rotate through different authentication profiles within the same provider before falling back to a different provider. This can significantly reduce interruptions caused by temporary quota limits, making your agent more resilient.

Stability Improvements That Matter

Beyond new features, this release addresses a number of stability issues reported by users. These fixes are aimed at making OpenClaw more reliable in day-to-day use.

No More Gateway Hangs After Startup

Some users reported that the gateway would hang about a minute after startup following an upgrade. The cause was identified as a task registry maintenance sweep that was blocking the gateway event loop under synchronous SQLite pressure. This has been fixed, and the gateway now runs without these interruptions.

Model Switching Without Interruption

Switching models during an active conversation used to risk interrupting the current turn. Now, /model changes are queued behind any running tasks. The switch takes effect only after the current turn finishes, and subsequent messages use the new model. This means you can adjust model settings without breaking the flow of a conversation.

Better Handling of Media and Downloads

  • Discord: Inbound media attachments and stickers now go through the same idle-timeout and worker-abort path as other downloads. This prevents slow or stuck media fetches from hanging message processing.
  • Telegram: Non-idempotent sends now use a stricter safe-send path. The agent also properly handles 429 (too many requests) responses and respects retry_after backoff periods.
  • Telegram Local Bot API: MIME types are now preserved for absolute-path downloads. This ensures that local audio files still trigger transcription and other MIME-based processing.

Data and Configuration Integrity

We made several fixes to ensure that your data and configuration remain consistent across restarts:

  • OAuth credentials: For OpenAI Codex, plugin-refreshed OAuth credentials are now persisted to auth-profiles.json before being returned. This stops the refresh_token_reused loop that could occur after a restart.
  • Configuration migration: Deprecated channels.telegram.groupMentionsOnly is now automatically migrated to channels.telegram.groups["*"].requireMention on load. This prevents crashes when using legacy configuration files.
  • Memory indexing: Full reindexes triggered by session-start or watch no longer skip session transcripts. This ensures that restart-driven reindexes preserve session memory.

Plugin and Runtime Improvements

  • Plugin loading: Bundled channel plugins now load correctly from legacy channels.<id> configuration even under restrictive plugin allowlists. The openclaw doctor output has also been refined to warn only about real plugin blockers.
  • Runtime dependencies: The externalized runtime dependency staging introduced in 2026.3.31 is now fully restored. Bundled plugins retain their declared runtime dependencies across packed installs, Docker builds, and local staging.
  • LINE channel: The LINE channel now starts correctly after global npm installs, thanks to a fix in the packaged runtime contract.
  • MiniMax plugin: The bundled MiniMax plugin is now auto-enabled for API-key authentication, so image generation and other plugin capabilities load without manual allowlisting.

Frequently Asked Questions

How do I use the new /tasks command?
Simply type /tasks in any chat session. The agent will return a task board showing background tasks for that session, including status, recent activity, and fallback counts.

What does agents.defaults.params do?
This new configuration option lets you set global default parameters for your providers. For example, you can define a default temperature or max tokens that applies to all agents unless overridden. It simplifies configuration when you want consistent behavior across multiple sessions.

What should I know about the command approval changes?
If you previously experienced allow-always not sticking, or approvals that seemed inconsistent, upgrading should resolve those issues. We also recommend running openclaw doctor to check for conflicts between your global tools.exec settings and your exec-approvals.json rules.

I saw a “queue owner unavailable” error with ACPX. Is that fixed?
Yes. The session recovery logic has been improved. The system now replaces dead named sessions and resumes the backend session when ACPX provides a stable session ID. If the reported session ID is stale, the recovery process creates a fresh named session instead of failing.

Why did my Telegram forum topic approvals go to the root chat?
This was a bug in how topic-aware exec approvals were routed. It has been fixed. Approvals now stay in the originating topic rather than falling back to the root chat.

Updating to 2026.4.1

We recommend updating to this version to take advantage of the stability improvements and new features. Depending on your installation method, you can update with one of the following:

# For npm installations
npm update -g @openclaw/openclaw

# For Docker installations
docker pull openclaw/openclaw:latest

For other installation methods, please refer to the official documentation.

Final Thoughts

OpenClaw 2026.4.1 is a refinement-focused release. It strengthens the foundations of the platform—task management, cross-platform reliability, command execution security, and system stability—without introducing disruptive changes. Whether you use OpenClaw for personal automation, team collaboration, or as part of a larger development workflow, these updates are designed to make your experience more reliable and intuitive.

We are grateful to the community for the feedback, issue reports, and contributions that shaped this release. Your input continues to drive OpenClaw forward.