update
Jun 3, 2026
By Teun
v0.30.2 adds Cline CLI install, Qwen code integration
v0.30.2 adds launch-time prompts to show and install the Cline CLI, along with Qwen code integration and Codex launch configuration isolation. The release also tightens URL handling, updates llama.cpp, and adds fixes for model limits, token counting, and load-stall detection.
v0.30.2 ships with a set of launch, model, and security-related changes across the app. The release notes list new support for showing and auto-installing the Cline CLI, Qwen code integration, and isolated launch configuration for Codex, along with several fixes aimed at improving local model handling and troubleshooting.
According to the changelog, the launch flow now shows and can auto-install the Cline CLI in feature work merged by @hoyyeva in #16402. The same area also adds Qwen code integration in #15900, which suggests the launcher now has a clearer path for connecting that model or coding setup.
⚡ New to this?
This release matters because it changes how the app starts tools, connects to model backends, and handles content like markdown links. “llama-server” is the service that runs a model and answers requests, while “SSE ping comments” are keepalive messages used to show the server is still alive. These kinds of updates affect stability, compatibility, and security for people running AI tools locally.
🦞 OpenClaw angle
If you self-host agents or model servers, test this release against your existing launch profiles before rolling it into production. Pay attention to Codex, opencode, and llama-server setups, since this version changes launch configuration, local model limits, token counting, and stall detection. If your workflows parse markdown or process user-supplied links, re-check that path after the URL-handling hardening changes.
The release also isolates Codex launch configuration in #16437, a change that should help keep that setup from affecting other launch profiles. In a separate launch fix, @dhiltgen adjusted local model limits for opencode in #16425.
On the model-serving side, the release includes several llama and llama.cpp updates. The notes say cached prompt tokens are now included in llama-server counts in #16428, and llama-server load stalls can now be detected from output in #16427. The changelog also says SSE ping comments from llama-server are ignored in #16443, which points to cleaner handling of server health and status messages.
There is also an update to llama.cpp itself in #16426. Another model-related change adds laguna, described as a Poolside architecture, through a llama.cpp patch under llama/c in #16396, followed by a fix for the laguna patch build breakage in #16445.
The release includes hardware discovery work as well. According to the notes, Radeon 8060S iGPU is now allowed by default in discovery code in #16429. That should broaden the set of systems the app recognizes without extra configuration.
Security and reliability changes appear in the markdown handling code. The changelog says app markdown URL handling was hardened in #16380 and then hardened further in #16436. That kind of change typically means the app is being stricter about how it processes links found in markdown content.
Finally, the release adds troubleshooting help and documentation updates. @dhiltgen added logging of template details to aid troubleshooting in #16403, and the docs were updated to configure the Hermes desktop app in #16440. The full changelog points back to v0.30.0 through v0.30.2-rc0 for the complete list of changes.
