ClawMax v1.9.9 Enhances OpenClaw Agent Orchestration

What changed

ClawMax has released version 1.9.9, a stable update to its web orchestration layer for OpenClaw agents, teams, workflows, and templates. This release brings several key enhancements, including native integration for OpenRouter and xAI/Grok providers, expanding the range of AI models that can be utilized within the platform. A significant addition is the implementation of brokered agent-skill secret grants, which allows for more granular and secure management of sensitive information required by agents and their skills. The update also introduces improvements to mobile-safe workflow runs, ensuring a better user experience on various devices.

Further refinements in version 1.9.9 include clearer feedback mechanisms for long-running tools, making it easier for developers to monitor and debug complex operations. The release also ensures pinned OpenAI model compatibility, providing stability for existing integrations. This version retains the OpenClaw v2026.6.11 baseline and was promoted from the 1.9.9-test-rc6 image. Notably, browser-local secrets remain unavailable to agent chat by design; runtime use necessitates an explicitly assigned skill, an exact secret grant, and a registered broker action, with standard Google account passwords not supported.

Development is also progressing on the 2.0.0 line, with RC83 preparation focusing on strengthening isolated runtime validation and aiming for at least 75% branch coverage. The main branch now includes the public clawmax.ai/v2 plugin platform, enabling plugins to contribute various components like pages, APIs, and data through a domain-neutral host contract. Public 2.0 products are slated to include AI scoring, Lifecycle and Review plugins for agent/workflow inspection, and curated integrations for Gmail and Microsoft 365/Outlook. An Activity Export contract for partner event integrations, such as Digo, is also planned, which will be consent-gated and clearly indicate its destination without blocking agent execution.

Previous highlights in the 1.9.x series include version 1.9.8 with federated skill search and safe DocHub bulk file management, 1.9.7 which provided the first fully validated OpenClaw v2026.6.11 release across multiple environments and introduced automatic workflow session-conflict recovery, and 1.9.5 featuring performance baseline artifacts and direct custom-skill ZIP upload/import. Earlier versions like 1.9.3 addressed chat archive correctness and template mobile overflow, while 1.9.2 improved diagnostics and safer local skill import. Version 1.9.1 focused on test backfilling and broader regression coverage.

Why it matters for builders

This release empowers builders by expanding the ecosystem of supported AI providers and enhancing the security posture for managing secrets. The introduction of brokered secret grants and the continued focus on secure runtime operations provide developers with more control and confidence when integrating sensitive data into their agent workflows. Furthermore, the ongoing development towards version 2.0, with its plugin platform and enhanced integration capabilities, signals a future where ClawMax will offer even greater extensibility and interoperability for complex AI agent systems.

Practical impact

Developers using ClawMax can now integrate with OpenRouter and xAI/Grok models directly, potentially unlocking new capabilities or cost efficiencies. The improved secret management system, particularly brokered grants, offers a more secure way to handle API keys and credentials for agent skills, reducing the risk of exposure. Builders should explore the updated documentation for configuring these new providers and understanding the brokered secret grant mechanism. For those planning future development, the advancements in the 2.0 release, such as the plugin platform, suggest opportunities for creating custom extensions and deeper integrations with third-party services like Gmail and Microsoft 365.

Caveats and source limits

The source material indicates that browser-local secrets are intentionally not available to agent chat by design, requiring explicit grants for runtime use. While version 1.9.9 is a stable release, the 2.0.0 line is still under development, with RC85 validation and installed-instance acceptance pending for its launch. Specific details on performance benchmarks or independent validation of the new provider integrations are not provided in the source. The source also notes that browser-local values are not a substitute for proper server-side secret management and should be treated as locally accessible if a browser profile is shared.