By devasher · Edited by Nominiclaw
This beta release introduces significant performance optimizations for model listing, enhanced security for plugin discovery, and critical fixes for macOS LaunchAgent stability and Codex runtime behavior.
/models command and model pickers have seen a massive performance boost, dropping from ~20 seconds to ~5ms. This was achieved by pre-warming the provider auth-state map at gateway startup, eliminating redundant discovery sweeps on the hot path.agents, setup, onboard, and configure now lazy-load their respective action runtimes. This reduces startup time and memory overhead (RSS) significantly when users are simply seeking help documentation.SUBAGENTS.md or AGENTS.md only). This prevents delegated workers from wasting tokens on full persona or user context they don't need.spawn_agent completions were not proactively delivered back to the parent session, which previously required a manual user poke.timeout, runtime_unavailable) instead of generic internal_error, allowing for better programmatic handling by hooks and plugins.config.plugins.allow to be executed during discovery.mode=reload restarts where launchctl bootstrap could fail with an I/O error if the async unload hadn't completed. The system now polls for unload completion and falls back to kickstart.createToolbarButton function.web_search, including active-agent auth threading and new model aliases like grok-build-0.1.web_search calls with freshness filters failed due to sub-second timestamp precision. Timestamps are now stripped to second-precision to meet Gemini API requirements.paste-token produced the wrong credential shape. A new paste-api-key command has been added to handle this correctly.agentComponents.ttlMs (up to 24 hours) for component callback registries.isReasoning payloads to prevent model "thinking" text from leaking into visible messages./models command and faster CLI help access.plugins.allow array in your configuration.openclaw models auth paste-token --provider openai-codex may find their credentials invalid. You should migrate these by running:
openclaw models auth paste-api-key --provider openai-codex