Ai

OpenClaw v2026.4.23 Expands Image Generation and Sub-Agent Context Controls

OpenClaw v2026.4.23 Expands Image Generation and Sub-Agent Context Controls

OpenClaw v2026.4.23 delivers a meaningful update for AI workflow builders, centered on broader image-generation support, more flexible sub-agent context handling, and finer-grained control over generation and memory behavior. The release makes version-specific improvements that reduce setup friction for image features, especially by enabling openai/gpt-image-2 through Codex OAuth without requiring a direct OpenAI API key, while also extending image_generate support to OpenRouter-backed image models.

What Changed

The most visible change in v2026.4.23 is the expansion of image-generation support across providers. OpenAI image generation and reference-image editing now work through Codex OAuth, allowing openai/gpt-image-2 to run without an OPENAI_API_KEY. OpenRouter also gains image generation and reference-image editing through the same image_generate tool when an OPENROUTER_API_KEY is present. This release also adds provider-supported quality and output-format hints, along with OpenAI-specific options for background, moderation, compression, and user hints.

On the agent orchestration side, OpenClaw adds optional forked context for native sessions_spawn runs. Child agents can now inherit requester transcript context when that is useful, while isolated sessions remain the default. The release also introduces optional per-call timeoutMs support for image, video, music, and TTS generation tools, making it easier to extend provider request windows only for heavier generation tasks instead of changing defaults globally.

Memory behavior also gets a host-tuning improvement. Local embeddings now support configurable memorySearch.local.contextSize, with a default of 4096, so constrained systems can reduce context size without patching the memory host itself. In addition, bundled Pi packages were updated to 0.70.0, and OpenClaw now uses Pi's upstream gpt-5.5 catalog metadata for OpenAI and OpenAI Codex, while keeping only local gpt-5.5-pro overrides.

Why It Matters

This version matters because it removes practical friction from multimodal workflows. Teams already authenticated through Codex OAuth can now access OpenAI image generation without provisioning a separate OpenAI API key, while OpenRouter users gain a clearer path to image-capable models through the native tooling layer. That makes OpenClaw more flexible for mixed-provider deployments.

The new forked-context behavior for sessions_spawn is also important for agent design. It gives builders a middle ground between fully isolated child sessions and manually stuffing extra context into prompts, which should improve handoffs for complex multi-agent flows while preserving isolation as the safe default. Combined with per-call generation timeouts and configurable local embedding context size, v2026.4.23 gives operators more control over performance, reliability, and resource usage on real-world hosts.

Official Source: https://github.com/openclaw/openclaw/releases/tag/v2026.4.23