The 2026.5.28 release is a big one. New models, a completely new orchestration layer, a refreshed iOS app, and the most reliable runtime recovery we’ve ever shipped. Let’s get into it.
Claude Opus 4.8 and New Provider Support
OpenClaw now supports Claude Opus 4.8 — the latest from Anthropic. If you’ve been waiting for the newest Opus model, it’s ready to go. Just update your model configuration and point your agents at it.
But that’s not all on the provider front:
- Fal Krea image models: New image generation schemas for Fal’s Krea models, giving you more creative AI image options.
- NVIDIA featured model catalogs: Browse and use NVIDIA’s featured models directly from OpenClaw.
- MiniMax streaming music: MiniMax now supports streaming music responses — real-time audio generation.
- Provider-backed voice model catalogs: Voice models from providers are now cataloged and selectable, making it easier to pick the right voice for your agents.
What This Means For You
More models means more options. Claude Opus 4.8 is a significant capability jump for complex reasoning tasks. The Fal Krea integration opens up new creative workflows. And the voice model catalogs mean you’re no longer locked into a single TTS provider — you can pick the voice that fits your use case.
If you’re building multi-modal agents that combine text, image, and voice, this release gives you the most provider flexibility yet.
Workboard: Multi-Agent Orchestration
This is the headline feature. Workboard is a new orchestration layer for managing multi-agent workflows. It provides:
- Agent coordination tools: Plan, assign, and track work across multiple agents simultaneously.
- Run tracking: See the status of every agent run, including dependencies, outputs, and failures.
- Multi-agent planning: Define complex workflows where agents hand off work to each other in structured sequences.
Workboard isn’t just a task queue. It’s an orchestration framework that understands agent capabilities, dependencies, and state. You define the plan, and Workboard coordinates the execution.
What This Means For You
If you’ve been running multiple agents and manually coordinating their outputs — copy-pasting results between sessions, tracking progress in a spreadsheet, or just hoping everything comes together — Workboard eliminates that overhead.
Think of it as a project manager for your AI agents. You define the workflow, assign agents to steps, and Workboard handles the execution, tracking, and error recovery. For complex tasks like multi-step research, content pipelines, or code generation + review workflows, this is transformative.
iOS Pro App Refresh
The iOS app got a complete refresh for Pro users:
- Pro Command, Chat, Agents, and Settings: Full-featured interface for managing your OpenClaw agents from your phone.
- Hosted push relay defaults: Push notifications work out of the box with the hosted relay — no custom configuration needed.
- Realtime Talk playback: The Talk tab now plays audio in realtime, wired directly to gateway sessions. No more waiting for the full response before hearing anything.
- Improved onboarding: Getting started with the iOS app is smoother, with better permission flows and session picker behavior.
What This Means For You
If you’re an iOS user, this is the update you’ve been waiting for. The app is now a first-class client for your OpenClaw agents. Realtime Talk playback means you can have actual voice conversations with your agents on mobile. The hosted push relay means notifications just work — no server-side setup required.
GitHub Copilot Agent Runtime
The GitHub Copilot integration is now official:
- @openclaw/copilot plugin: Externalized as the official plugin with npm and ClawHub publish metadata. Install it, configure it, and your Copilot agent runtime is ready.
- Codex Supervisor plugin path: For delegated Codex workflows, the Supervisor plugin path gives you structured control over how Codex sessions are managed.
What This Means For You
If you’re using GitHub Copilot with OpenClaw, the setup is now standardized and supported as a first-party plugin. No more custom configuration or hoping your hack still works after an update. The npm package makes installation clean, and the ClawHub metadata means it’s discoverable and versioned properly.
Tokenjuice Plugin
Tokenjuice is now an official @openclaw plugin. If you’ve been tracking token usage across your agents, this is the standardized way to do it. Install from npm, configure your tracking preferences, and get consistent token accounting across all your agents.
Rock-Solid Runtime Recovery
This section might be the most important part of the release, even though it’s the least glamorous:
- Subagent cwd/workspace separation: Subagents keep their working directory and workspace separation intact, even through crashes and restarts.
- Hook context stays prompt-local: Hook context doesn’t leak across prompts. Each prompt gets its own isolated context.
- Session lock management: Session locks release on timeout abort while live OpenClaw locks survive cleanup. No more deadlocks from crashed sessions.
- Stale restart avoidance: Continuations from stale restarts are avoided. Your agents don’t pick up abandoned state.
- Codex failure isolation: Codex app-server and helper failures no longer tear down shared runtime state. One Codex crash doesn’t take down the whole runtime.
What This Means For You
Reliability. Pure and simple. If you’re running OpenClaw in production — especially with multiple agents, subagents, and Codex sessions — these fixes prevent the cascading failures that cause mysterious downtime.
The session lock management alone is worth the upgrade. If you’ve ever had a gateway that wouldn’t restart because a session lock was stuck, that problem is gone. The failure isolation for Codex means that even if one agent’s Codex session crashes, it doesn’t affect any other agent on the same runtime.
Channel Delivery Improvements
Every major channel got delivery reliability improvements:
- Plugin hooks: Safer outbound delivery through plugin hooks.
- Matrix: Room IDs handled correctly.
- iMessage: Reactions and approvals work reliably.
- Slack: Final replies deliver correctly every time.
- Discord: Tool warnings recovered gracefully.
- WhatsApp: Profile auth roots validated properly.
- Telegram: Polling is more robust.
- Microsoft Teams: Service URL trust checks are stricter.
- Runtime-config message actions: Message actions from runtime config are handled correctly.
What This Means For You
Fewer dropped messages, fewer mysterious delivery failures, and more reliable bot behavior across every channel you use. If you’ve been seeing intermittent delivery issues on any channel, this release likely fixes them.
Stricter Input Validation
Garbage in, garbage out — unless you reject the garbage at the door. This release adds earlier validation for:
- Browser tool timeouts and viewport/tab indices
- Gateway ports
- Cron retry handling
- Discord component IDs
- Schema array references
- Telegram callback pages
- Channel progress callbacks
Malformed values are rejected earlier, which means clearer error messages and less time debugging mysterious downstream failures.
What This Means For You
If you’re building integrations or custom tooling on top of OpenClaw, you’ll get better error messages when something is wrong with your inputs. The validation catches problems at the API boundary instead of letting them propagate into the runtime where they cause confusing errors.
PDF Improvements
Two significant PDF improvements:
- ClawPDF extraction: Better PDF text extraction for document processing agents.
- Encrypted PDF support: OpenClaw can now handle encrypted PDFs, opening up use cases that were previously blocked.
- MCP structured content: Structured content in MCP (Model Context Protocol) tool results is now supported in agent tool responses.
Policy Additions
New policy features for governance and compliance:
- Policy comparison: Compare policies to see what changed between versions.
- Ingress-channel conformance: Check that incoming channels conform to your defined policies.
- Sandbox-posture conformance: Verify that your sandbox configurations meet your compliance requirements.
What This Means For You
If you’re managing OpenClaw in an environment with compliance requirements, these policy tools give you auditable control over what your agents can do and how they’re configured. Policy comparison alone is valuable for change management — you can see exactly what changed before applying a new policy.
Discord Progress Improvements
Discord now shows commentary in progress drafts. When an agent is working on a long task, the in-progress draft in Discord shows useful context about what’s happening. No more staring at a blank “thinking…” message — you can see the agent’s reasoning in real time.
Performance Optimizations
The usual round of performance improvements, this time focused on reducing repeated work in hot paths:
- Install records cached
- Config JSON parsing reduced
- Tool search catalogs cached
- Session stores optimized
- Manifest model rows cached
- Auto-enabled plugin config cached
- Browser tokens cached
- Viewer assets cached
- Release-split external plugin packages handled more efficiently
What This Means For You
Faster everything. Less CPU, less memory, more throughput. The cumulative effect of these caching improvements is significant — especially on gateways handling many concurrent sessions.
Build In Public: 5.28 Release Coverage
The Build In Public YouTube channel released a comprehensive video covering the 5.28 update. Check it out for a walkthrough of the major features:
Upgrade Guide
Update with npm:
npm update -g openclawOr install fresh:
npm install -g openclawThen restart your gateway:
openclaw gateway restartNo breaking configuration changes. Existing plugins and agent configurations should work without modification.
—
The 2026.5.28 release is a milestone. New models, new orchestration, new mobile experience, and the most reliable runtime we’ve ever shipped. Don’t sleep on this one — update today.
