[Release] openclaw/openclaw v2026.4.23-beta.5: OpenClaw 2026.4.23 beta 5

OpenClaw v2026.4.23-beta.5 introduces significant enhancements to image generation across multiple providers (OpenAI, OpenRouter), adds optional forked context for subagents, implements per-call timeout support for generation tools, and includes configurable local embedding context sizes. The release fixes numerous issues across Codex harness routing, media handling, provider authentication, and UI persistence, while updating Pi dependencies to v0.70.0 and improving debug logging for harness selection decisions.

Key Points

• •Image generation now works without API keys via Codex OAuth for OpenAI and through OpenRouter's image_generate tool, with support for quality hints, output formats, and provider-specific parameters
• •Subagents can optionally inherit requester transcripts through forked context while maintaining isolated sessions as default, with prompt guidance and context-engine hook metadata
• •Per-call timeoutMs support added for image, video, music, and TTS generation tools to extend provider request timeouts on demand without global changes
• •Local embeddings now support configurable memorySearch.local.contextSize (default 4096) for tuning on constrained hosts without code patching
• •Codex harness routing improved with structured debug logging for embedded harness selection and fallback decisions, separate from simplified /status endpoint
• •Fixed critical issues: native request_user_input prompts routed to originating chat, context-engine assembly failures redacted before logging, Windows npm-installed codex.cmd shims resolved through PATHEXT
• •Media handling enhanced: Telegram markdown image syntax parsed into media payloads, image attachments preserved for text-only models via media refs, WhatsApp media normalization unified
• •Provider authentication hardened: OpenAI image generation auth routing improved, OpenRouter vision prompts reordered for non-empty responses, private-network SSRF opt-in honored for Gemini and LocalAI
• •Control UI improvements: assistant-generated images persisted as authenticated managed media with paired-device token support, Stop-button aborts queued across Gateway reconnects
• •Memory and agent fixes: stale QMD collections recreated on startup, OpenAI capacity failures surfaced with model-switch hints, transcript replay synthesis disabled for OpenAI/Codex while preserved for other providers

Workflow Diagram

Concepts