OpenClaw 2026.4.14: Complete Breakdown

Major Features

GPT-5.4 Pro Forward Compatibility

OpenClaw now includes forward-compat support for gpt-5.4-pro, including Codex pricing, limits, and list/status visibility. This means when OpenAI releases GPT-5.4 Pro, your OpenClaw deployment will recognize it immediately without requiring an emergency update.

Telegram Forum Topics Enhancement

Human-readable topic names now surface in agent context, prompt metadata, and plugin hook metadata. Previously, forum topics appeared as opaque numeric IDs. Now agents understand they are in "Bug Reports" or "Feature Requests" rather than "topic_42".

// Topic names persist across restarts
{
  "channels": {
    "telegram": {
      "groups": {
        "-1001234567890": {
          "topics": {
            "5": "Bug Reports",
            "12": "Feature Requests"
          }
        }
      }
    }
  }
}

Critical Fixes for GPT/Codex Users

Embedded Run Timeout Fix

The configured embedded-run timeout now properly flows into the global undici stream timeout tuning. Previously, slow local Ollama runs would hit the default stream cutoff instead of respecting your operator-set timeout. This caused premature failures on long-running local model operations.

Codex Provider Catalog Validation

The apiKey field now appears in Codex provider catalog output. This fixes a bug where the Pi ModelRegistry validator would reject the entry and silently drop all custom models from every provider in models.json. Your custom model configurations now persist correctly.

Minimal Thinking Mapping

OpenAI's minimal thinking setting now maps correctly to OpenAI's supported low reasoning effort for GPT-5.4 requests. Previously, embedded runs would fail request validation when minimal thinking was specified.

Legacy Alias Canonicalization

The legacy openai-codex/gpt-5.4-codex runtime alias now canonicalizes to openai-codex/gpt-5.4 while still honoring alias-specific and canonical per-model overrides. This eliminates confusion when using older configuration references.

Security Hardening

Dangerous Config Protection

The gateway tool now rejects config.patch and config.apply calls when they would newly enable any flag enumerated by openclaw security audit. This includes flags like:

• dangerouslyDisableDeviceAuth
• allowInsecureAuth
• dangerouslyAllowHostHeaderOriginFallback
• hooks.gmail.allowUnsafeExternalContent
• tools.exec.applyPatch.workspaceOnly: false

Already-enabled flags pass through unchanged, so non-dangerous edits in the same patch still apply. Direct authenticated operator RPC behavior is unchanged.

Slack Interactive Security

The configured global allowFrom owner allowlist now applies to channel block-action and modal interactive events. Interactive triggers can no longer bypass the documented allowlist intent in channels without a users list. Open-by-default behavior is preserved when no allowlists are configured.

Browser & Media Fixes

Browser SSRF Policy Restoration

Hostname navigation is restored under the default browser SSRF policy while keeping explicit strict mode reachable from config. Managed loopback CDP /json/new fallback requests stay on the local CDP control policy. This fixes regressions where browser follow-up fixes blocked normal navigation or self-blocked local CDP control.

Local Browser Reconnection

Loopback CDP readiness checks are now reachable under strict SSRF defaults. OpenClaw can reconnect to locally started managed Chrome without misclassifying a healthy child browser as "not reachable after start".

Image/PDF Tool Normalization

Configured provider/model refs are now normalized before media-tool registry lookup. This fixes cases where image and PDF tool runs rejected valid Ollama vision models as unknown because the tool path skipped the usual model-ref normalization step.

Google Image Generation Fix

A trailing /openai suffix is now stripped from configured Google base URLs only when calling the native Gemini image API. This fixes 404 errors on Gemini image requests without breaking explicit OpenAI-compatible Google endpoints.

Agent & Subagent Improvements

Subagent Registry Fix

The subagent registry lazy-runtime stub now emits on the stable dist path that both source and bundled runtime imports resolve. This fixes ERR_MODULE_NOT_FOUND errors at runtime when using runtime: "subagent".

Subagent NPM Build Fix

The dist/agents/subagent-registry.runtime.js file is now shipped in npm builds. Previously, runtime: "subagent" runs would stall in queued state after the registry import failed.

Context Engine Compaction

Engine-owned sessions now compact from the first tool-loop delta, and ingest fallback is preserved when afterTurn is absent. Long-running tool loops can now stay bounded without dropping engine state.

Tool Retry Logic

Streamed unknown-tool retries are now only marked as counted when a streamed message actually classifies an unavailable tool. Incomplete streamed tool names no longer reset the retry streak before the final assistant message arrives.

Cron & Scheduling Fixes

Cron Next-Run Repair

The scheduler no longer invents short retries when cron next-run calculation returns no valid future slot. A maintenance wake stays armed so enabled unscheduled jobs recover without entering a refire loop.

Error Backoff Preservation

The active error-backoff floor is now preserved when maintenance repair recomputes a missing cron next-run. Recurring errored jobs no longer resume early after a transient next-run resolution failure.

Memory & Context Fixes

Active Memory Path Change

Recalled memory now moves onto the hidden untrusted prompt-prefix path instead of system prompt injection. The visible Active Memory status line fields are now labeled, and the resolved recall provider/model appears in gateway debug logs. This ensures trace/debug output matches what the model actually saw.

Ollama Embedding Restoration

The built-in ollama embedding adapter is restored in memory-core. Explicit memorySearch.provider: "ollama" works again. Endpoint-aware cache keys ensure different Ollama hosts do not reuse each other's embeddings.

Memory/QMD Collection Fix

Legacy lowercase memory.md is no longer treated as a second default root collection. QMD recall no longer searches phantom memory-alt-* collections, and builtin/QMD root-memory fallback stays aligned.

Channel-Specific Fixes

WhatsApp Media Encryption

Installed Baileys media encryption writes are now patched during OpenClaw postinstall. The default npm/install.sh delivery path waits for encrypted media files to finish flushing before readback, avoiding transient ENOENT crashes on image sends.

BlueBubbles Cache Refresh

The Private API server-info cache now lazy-refreshes on send when reply threading or message effects are requested but status is unknown. Sends no longer silently degrade to plain messages when the 10-minute cache expires.

Feishu Allowlist Canonicalization

Allowlist entries are now canonicalized by explicit user/chat kind. Repeated feishu:/lark: provider prefixes are stripped, and opaque Feishu IDs are no longer folded to lowercase. Allowlist matching no longer crosses user/chat namespaces or widens to case-insensitive ID matches.

TTS Voice Note Fix

OpenClaw temp voice outputs now persist into managed outbound media and pass through reply-media normalization. Voice-note replies no longer silently drop.

Upgrade Recommendations

Upgrade Command

# Via CLI
openclaw update

# Or via agent prompt
"Please update OpenClaw to version 2026.4.14"

Post-Upgrade Verification

1. Run openclaw doctor to verify installation health
2. Test a GPT-5.4 embedded run with reasoning enabled
3. Verify Telegram forum topics show human-readable names
4. Confirm subagent spawning works without module errors
5. Check browser automation starts correctly

Release date: April 14, 2026 · 84 community reactions · 44 mentions · 50+ fixes
Full changelog: github.com/openclaw/openclaw/releases/tag/v2026.4.14