OpenClaw 2026.5.5 Release: Cross-Platform Messaging, Gateway Stability, and Developer Experience Fixes Explained

OpenClaw officially shipped v2026.5.5 on May 5, 2026. This isn’t a feature-packed release — there are no new modules, no flashy UI redesigns. But if you’re running OpenClaw to manage multi-platform AI agent deployments, the sheer volume of fixes in this update touches nearly every pain point you’d encounter in day-to-day operations.

From garbled Feishu topic routing to Discord heartbeat disconnects to TUI sessions resurrecting weeks-old messages on first boot — these long-standing issues have all been addressed. This article breaks down the fixes by functional module so you can quickly understand the real-world impact of this release.

1. Multi-Platform Messaging Channel Fixes

One of OpenClaw’s core value propositions is connecting AI agents to multiple instant-messaging platforms. This release delivers important stability improvements across several messaging channels.

1.1 Feishu: Topic Session Routing Fix

Feishu users had been running into a problem where missing native topic starter thread IDs caused session routing to break. When this happened, a user’s first message and subsequent replies would land in different sessions, causing the agent to “forget” the earlier conversation context.

The v2026.5.5 fix hydrates any missing native topic starter thread IDs before session routing takes place. As a result, both first turns and follow-ups stay within the same topic session, preserving conversation continuity.

1.2 LINE: DM Policy Validation Hardening

The LINE channel had a subtle configuration trap: when users set dmPolicy: "open" without configuring a wildcard allowFrom, webhook DM messages would be silently swallowed. The system would return an acknowledgement (making it look like everything succeeded), but the messages never actually entered inbound processing.

This release changes the behavior entirely — such configurations now trigger an explicit validation failure, surfacing the problem instead of silently dropping messages in the background.

1.3 Telegram and Codex: Progress Draft Deduplication

In the Telegram and Codex integrations, progress drafts were being duplicated when the agent used the message tool. The same progress line would appear twice — once as an “item draft” and once as a “tool draft.”

After the fix, only the message-tool-only progress draft remains visible, and native Codex tool progress renders once per tool instead of duplicating lines.

1.4 Discord: Heartbeat Mechanism and Command Routing

Discord received two significant fixes in this release:

Heartbeat timeout calculation fix: Previously, heartbeat ACK timeouts were measured from when the heartbeat request was sent, but initial heartbeats could be delayed. This caused premature timeout detection while the channel was still awaiting readiness, triggering unnecessary reconnect loops. The fix measures timeouts from the actual heartbeat send time, eliminating false positives.

Plain-text command routing: Commands like /steer that use plain text were being silently dropped before the agent session could see them. Now these commands go through the normal authorization and mention gate, ensuring the agent receives and processes them.

Additionally, Discord streaming replies now display reasoning text directly in progress drafts instead of showing a blank “Reasoning” status line.

1.5 Slack: Error Log Completeness

The Slack Socket Mode reconnect logs were losing critical context — SDK error details and structured Slack API fields were being collapsed into a bare unknown error message.

After the fix, reconnect logs preserve full error context and API field information, helping developers pinpoint the root cause of startup failures quickly.

1.6 WhatsApp: Responsiveness Optimization

The WhatsApp channel experienced reply delays when the Gateway event loop was under heavy load. The previous approach would stop all local TUI clients, which was too aggressive.

The refined logic now only stops a local TUI client when it has been verified as the source of Gateway event loop degradation and reply delays, avoiding collateral damage to healthy connections.

1.7 Matrix: Approval Message Retry Mechanism

Approval prompts sent through the Matrix channel previously had no retry mechanism. If a Matrix send experienced a transient failure, the pending approval prompt would simply be stranded — the user would never receive the confirmation request.

The new version adds up to 3 retries with a short backoff strategy, significantly reducing the chance of approval workflow interruptions from transient send failures.

2. AI Model Provider Compatibility Fixes

OpenClaw supports integration with multiple AI model providers. This release delivers critical compatibility fixes for xAI and Fireworks.

2.1 xAI (Grok Models): Reasoning Parameter Compatibility

This was a direct usability issue. OpenClaw was sending OpenAI-style reasoning effort controls to xAI’s native Grok Responses models, but xAI’s API doesn’t support these parameters. This caused xai/grok-4.3 to fail in live Docker/Gateway runs with Invalid reasoning effort.

The fix operates on two levels:

  • Stop sending incompatible parameters: OpenAI-style reasoning effort controls are no longer sent to native Grok Responses models.
  • Lock the built-in configuration: The bundled xAI thinking profile is clamped to off, preventing unsupported reasoning levels from being sent to Grok models during live Gateway runs.

2.2 Fireworks (Kimi Models): Thinking Parameter Fix

The Fireworks provider’s Kimi models had a similar issue: K2.5 and K2.6 models could receive reasoning* parameters during manual model switches, and Fireworks would reject them.

After the fix, Kimi models are exposed as thinking-off-only, and K2.5/K2.6 requests always use thinking: disabled, preventing parameter rejection errors.

2.3 Auth Profiles: Format Rejection and Cooldown Mechanism

When a model name wasn’t supported, the system previously put the entire provider into cooldown. This meant other models under the same provider couldn’t be tried through fallback profiles either.

The fix ensures format-level rejections no longer trigger provider cooldown, so fallback profiles can still be attempted.

3. Control UI Improvements

The Control UI is OpenClaw’s web-based management interface. This release delivers multiple improvements in session management, performance, and chat experience.

3.1 Session Management Enhancements

Compaction count display: The compaction count now appears as a compact N Checkpoint(s) disclosure. Expanding it reveals modern checkpoint history cards in responsive table layouts.

Agent runtime visibility: The Sessions table now displays each session’s agent runtime and supports filtering by runtime labels, matching the wording used in the Agents panel.

Session creation hook fix: The documented /new command and lifecycle hooks were being triggered during SDK parent-session creation, causing session-memory and custom hook capture to malfunction. After the fix, these hooks fire only for explicit Control UI session creation.

3.2 Performance Optimization

When history payloads or channel probes are slow, the chat and channel tabs were previously freezing. The improvements include:

  • Chat and channel tabs remain responsive during data loading
  • Channel status is labeled as “partial” when data is incomplete
  • Slow chat/config render timings are recorded in the event log

3.3 Chat History Persistence

A usability issue affected daily workflows: when the same transcript turn contained both assistant progress text and tool-use metadata, reloading chat.history would cause the progress text to vanish after the next user message.

The fix ensures persisted assistant progress text remains visible even when tool-use metadata is present in the same turn.

4. Gateway Stability Fixes

The Gateway is OpenClaw’s core communication layer. This release delivers extensive stability improvements.

4.1 Shutdown and Restart Flow

Delayed maintenance task cancellation: Previously, delayed post-ready maintenance tasks were not cancelled during Gateway shutdown. After quick restarts, maintenance/cron tasks could be started redundantly, creating orphaned background timers. The fix cancels delayed maintenance during close and suppresses maintenance/cron startup after quick restarts.

Structured shutdown reporting: Shutdown warnings and HTTP close timeout warnings are now reported through ShutdownResult, preserving lifecycle hook hardening.

4.2 Health Status Detection

Fast repeated health/status samples were previously marking the event loop as “degraded” based solely on CPU/utilization data. The fix requires the Gateway to accumulate a sustained sampling window before making degradation judgments, reducing false positives.

Additionally, openclaw gateway status --deep now displays recent supervisor restart handoffs (including JSON details), reporting clean service-managed restarts as restart handoffs rather than opaque stopped-service diagnostics.

4.3 HTTP Route Optimization

  • Media handling optimization: Unrelated HTTP requests no longer trigger media sidecar handling, reducing processing overhead for non-media requests.
  • Disabled route responses: Disabled OpenAI-compatible routes now return 404 immediately instead of waiting for lazy-loaded media sidecar components.
  • Model catalog caching: Empty read-only model catalog results are cached until the next reload, preventing TUI and control-plane refresh loops from hammering plugin metadata reads when no usable models are discovered.

4.4 OpenAI-Compatible Streaming Output

An issue affecting API clients was that the initial assistant-role SSE chunk in streaming chat completions wasn’t being sent promptly. When agent cold-start setup took a long time, /v1/chat/completions clients could receive a bodyless 200 response and hang until their idle timeout fired.

The fix sends the assistant-role SSE chunk as soon as streaming chat-completion headers are accepted. The initial chat stream chunk flush logic has also been corrected, ensuring first-token streaming is visible and not delayed behind later chunks.

5. CLI and TUI Experience Improvements

5.1 TUI Fixes

Startup flow optimization: The TUI previously used a generic CLI respawn wrapper for interactive launches, wouldn’t exit cleanly on terminal loss, and could restore heartbeat sessions as the remembered chat session. This caused stale heartbeat history and orphaned openclaw-tui processes on first boot.

After the fix:

  • Interactive launches skip the generic CLI respawn wrapper
  • Terminal loss triggers a clean exit
  • Heartbeat sessions are no longer restored as remembered chat sessions

Session picker optimization: The session picker now bounds itself to recent rows and uses exact lookup-style refreshes for the active session, preventing dusty stores from making TUI hydrate weeks-old transcripts before becoming responsive.

5.2 CLI Status and Session Display

  • openclaw status session rows now show the selected agent runtime/harness, matching the /status runtime line.
  • The openclaw sessions table similarly displays agent runtime information.

5.3 CLI Update and Channel Commands

Dev-channel updates: The preflight lint for openclaw update --channel dev is now opt-in and constrained when enabled, preventing Ubuntu hosts from rolling back otherwise-good main commits due to OOM kills or failed parallel oxlint shards.

Channel help command: The openclaw channels parent-help command now skips config loading, proxy setup, channel-option catalog, banner config, and plugin startup bootstrap, exiting promptly after printing help.

Gateway command improvements: Non-TTY stdin is paused after CLI command completion; openclaw agent no longer falls back to embedded mode after Gateway request/auth failures, allowing parent help commands to exit cleanly.

5.4 Session Cleanup

sessions cleanup now prunes old unreferenced transcript files, compaction checkpoints, and trajectory artifacts, preventing Gateway restart or crash orphans from accumulating indefinitely outside sessions.json.

6. Plugin System Fixes

6.1 Plugin Sync and Updates

Sync during host updates: Installed official npm and ClawHub plugins (such as Codex, Discord, WhatsApp, and diagnostics plugins) are now synced during host updates, even when disabled or previously exact-pinned. Third-party plugin pins are preserved.

npm root state repair: Before plugin installs, the system repairs stale managed npm-root openclaw peer packages, preventing old core package-lock state from downgrading beta-channel official plugin updates.

Post-install peer links: After shared-root npm installs, updates, and uninstalls, the system reasserts managed npm plugin openclaw peer links, ensuring that mutating one plugin doesn’t leave previously installed SDK-using plugins unable to resolve openclaw/plugin-sdk/*.

Corrupt record tolerance: When corrupt managed plugin records are encountered during updates, core package updates can still complete and report the plugin repair path.

6.2 Plugin Diagnostics

When a TypeScript package has only source code without compiled runtime output, the warning message now explains that this is a publisher packaging issue and directs users to update/reinstall or disable/uninstall the plugin, making the warning actionable.

7. Agent Functionality Fixes

7.1 Context Engine

Hidden OpenClaw runtime-context custom messages were leaking into the context engine’s assemble, afterTurn, and ingest hooks, causing transcript reconstruction plugins to see non-conversation messages.

The fix ensures these hooks only process genuine conversation messages, providing cleaner input for transcript reconstruction plugins.

7.2 Generated Media Handling

Attachment-style message tools: When generated files have already been uploaded, attachment-style message tool actions are now treated as completed chat sends, preventing duplicate fallback media posts.

Async generation deduplication: While an announce-agent run is still pending, the system no longer triggers direct generated-media completion fallback, preventing async video and music completions from producing duplicate raw media messages.

7.3 Codex Media Paths

Codex app-server generated local images are now staged into managed media before Gateway display, resolving LocalMediaAccessError issues with Codex-home image paths while keeping Codex home out of the display allowlist.

7.4 Video Generation Parameter Compatibility

The video generation tool now accepts provider-specific aspect-ratio and resolution hints at the tool boundary. It normalizes 720P to MiniMax’s supported 768P and stops sending Google’s generateAudio parameter on Gemini video requests, allowing provider fallback to recover from model-specific parameter differences.

7.5 Agent Config Cleanup

The ambiguous legacy main agent directory helper has been removed from runtime paths. Model, auth, gateway, bundled plugin, and test helpers now resolve default/session agent directories through agents.list/agent-scope helpers, while the plugin SDK retains a deprecated compatibility export.

8. Doctor Diagnostic Tool Enhancements

openclaw doctor is OpenClaw’s built-in diagnostic and repair tool. This release enhances its capabilities across several areas.

8.1 Session Repair

Heartbeat-poisoned default main session store entries are now moved to recovery keys, and stale TUI restore pointers are cleared. This means doctor --fix can repair instances that were stuck on agent:main:main heartbeat history.

8.2 Gateway Diagnostics

openclaw doctor --deep now reports recent supervisor restart handoffs, using the installed service environment when available, making service-managed clean exits visible in guided diagnostics.

8.3 Codex Route Repair

Doctor can now repair legacy openai-codex/* routes and normalize them to openai/*. Coverage includes: primary models, fallbacks, heartbeat/subagent/compaction overrides, hooks, channel overrides, and stale session pins. When the Codex plugin is installed, enabled, contributes the codex harness, and has usable OAuth, agentRuntime.id: "codex" is selected; otherwise agentRuntime.id: "pi" is used.

8.4 Token Configuration Warnings

Doctor now warns when OPENCLAW_GATEWAY_TOKEN would shadow a different active gateway.auth.token source for local CLI commands, while avoiding false positives when config points at the same environment token.

9. Infrastructure and Security Hardening

9.1 Docker Container Security

The bundled docker-compose.yml now hardens the Gateway container with the following measures:

Security Measure Description
Remove NET_RAW capability Prevents raw network access
Remove NET_ADMIN capability Prevents network administration operations
Enable no-new-privileges Prevents process privilege escalation

9.2 Exec Approvals File Compatibility

On Windows systems, when rename-overwrite is rejected, writing exec-approvals.json now falls back to a guarded copy while preserving symlink, hard-link, and owner-only permission safeguards.

9.3 iOS Pairing Improvements

iOS pairing now allows setup-code and manual ws:// connections for private LAN and .local gateways while keeping Tailscale/public routes on wss://. In mixed-auth reconnects, explicit gateway passwords are preferred over stale bootstrap tokens.

9.4 Session Memory Hooks

Filename collision handling: Fallback memory filenames now include collision suffixes, preventing repeated /new or /reset captures within the same minute from overwriting earlier session archives.

Non-blocking reset: Reset memory capture now runs off the command reply path, and model-generated memory filename slugs are opt-in via llmSlug: true. This means /new and /reset no longer block WhatsApp and other message-channel reset replies on hook housekeeping or nested model calls.

10. Status Information Enhancements

Gateway Process and System Uptime

The /status command now displays compact Gateway process uptime and host system uptime, making restart checks and host-lifetime verification visible directly from chat.

11. Fix Summary by Module

For quick reference, here’s a consolidated overview of all fixes organized by module:

Module Fix Count Key Improvements
Messaging Channels 7 Feishu topic routing, LINE validation, Discord heartbeat & commands, Slack logging, WhatsApp responsiveness, Matrix retries
AI Providers 3 xAI reasoning parameter compatibility, Fireworks thinking parameter, auth cooldown mechanism
Control UI 4 Session management, performance optimization, chat persistence, runtime visibility
Gateway 6 Shutdown flow, health detection, HTTP routing, streaming output, media handling, model catalog caching
CLI / TUI 7 TUI startup, session picker, status display, update flow, session cleanup
Plugin System 5 Sync updates, npm repair, peer links, corrupt record tolerance, diagnostics
Agent 6 Context engine, generated media, Codex media, video parameters, config cleanup
Doctor 4 Session repair, Gateway diagnostics, Codex routing, token warnings
Infrastructure 4 Docker security, exec approvals, iOS pairing, session memory hooks
Status Info 1 Process and system uptime display

Frequently Asked Questions (FAQ)

Does this release include any new features?

No. v2026.5.5 is a pure bugfix release focused entirely on resolving issues in existing functionality. No new modules or features are introduced.

What do users of xAI Grok models need to do?

After upgrading to v2026.5.5, the Invalid reasoning effort error for xai/grok-4.3 in Docker/Gateway runs is automatically resolved. No configuration changes are needed — the system stops sending incompatible reasoning parameters to Grok models automatically.

What changed for Docker deployments?

The bundled docker-compose.yml removes NET_RAW and NET_ADMIN capabilities and enables no-new-privileges. If you use a custom Docker configuration, consider referencing the official docker-compose.yml to sync the security changes.

What exactly was the Feishu topic continuity issue?

When Feishu’s native topic starter thread ID was missing, a user’s first message and subsequent replies would be routed to different sessions. The agent would lose prior conversation context, effectively “forgetting” what the user had said before. After upgrading, the system automatically hydrates missing thread IDs before routing.

Why were old messages appearing on first TUI launch?

This was caused by heartbeat sessions being incorrectly restored as chat sessions. The TUI could load heartbeat history as if it were normal chat records, causing users to see messages from weeks ago. v2026.5.5 explicitly refuses to restore heartbeat sessions as chat sessions, fully resolving this issue.

What if my plugins were downgraded after a host update?

v2026.5.5 repairs stale npm-root peer package state before plugin installs and automatically syncs official plugin versions during host updates. If you previously experienced beta-channel plugins being downgraded by old package-lock state, the upgrade resolves this automatically.

What new issues can openclaw doctor --fix repair?

In addition to its previous capabilities, doctor --fix can now: repair heartbeat-poisoned main session store entries, clear stale TUI restore pointers, normalize legacy openai-codex/* routes, and detect Gateway token configuration conflicts.

Which providers are affected by the video generation parameter compatibility issue?

Primarily MiniMax and Google Gemini. MiniMax doesn’t support 720P resolution, so the system now automatically normalizes it to 768P. Google Gemini video requests no longer receive the unsupported generateAudio parameter.

Does this update require downtime?

The release notes don’t specify the upgrade procedure or downtime requirements. Refer to the official OpenClaw documentation for upgrade guidance.

Who contributed to this release?

v2026.5.5 involved numerous community contributors, including but not limited to: @joeyzenghuan, @Patrick-Erichsen, @bryce-d-greybeard, @NikolaFC, @ramitrkar-hash, @BunsDev, @Alex-Alaniz, @MilleniumGenAI, @draix, @googlerest, @vincentkoc, @frankekn, @shakkernerd, @VintageAyu, @edenfunf, @slideshow-dingo, @yelog, and others.

Final Thoughts

v2026.5.5 is a quintessential “infrastructure release” — it doesn’t chase the spotlight of new features but instead methodically addresses real-world operational issues across cross-platform messaging routing, model parameter compatibility, gateway stability, plugin system reliability, and more.

If you’re running OpenClaw in production to manage multi-platform AI agent deployments, this release deserves priority in your upgrade schedule. Before upgrading, consider running openclaw doctor to check your current instance state. After upgrading, run openclaw doctor --deep to confirm all fixes have taken effect.