Cline Desktop v0.0.35 Adds Linux Support, Fixes Plugin Slash Commands and Voice Input

Cline shipped desktop-v0.0.35 on September 24, bringing native Linux support to its desktop coding assistant for the first time. Each release now includes x64 .deb and .rpm packages alongside existing macOS and Windows builds. There is no AppImage yet.

The Linux build uses the native GTK file picker for the "Open folder…" action, and updates now download in the background and install only when the user selects "Restart now," removing surprise password prompts during updates.

Plugin slash commands fixed

Plugin-registered slash commands, such as /goal, previously got sent to the model as plain text instead of triggering the plugin's own handler. In v0.0.35, these commands now run correctly, display the plugin's reply, and only start a new conversation turn when the command requires one. Enabled plugin commands now surface in the slash menu, and skills/workflows in that menu pull from the conversation's own workspace, including git worktrees. Cline also fixed a bug where a single broken plugin would cause every slash prompt in the session to fail.

Diagnostics, voice input, and reasoning effort

Settings gains a new Diagnostics row. Clicking Export writes a single text file to the Downloads folder containing the app version, OS details, settings, recent sidecar and hub logs, and manifests of selected sessions. According to Cline, API keys, credential-shaped values, prompts, and the home directory path are stripped from the export, making it safe to attach to GitHub issues.

Voice input is working again with live, streaming transcription from OpenAI, Vercel AI Gateway, and ElevenLabs. If the network drops mid-recording, the app falls back to the browser's built-in recognizer and retries the configured provider on the next attempt. The model picker now labels which voice models support realtime transcription versus recording-based transcription.

Reasoning effort settings are now remembered per provider. Previously, reopening a session reset the thinking picker to "Low" and sent that setting with the next message regardless of prior choice; switching providers now correctly applies the effort level last set for that specific provider.

Reliability and error handling

Model lists for LiteLLM, Baseten, Hicap, Poolside, Ollama, and LM Studio now surface actual connection errors instead of showing an empty list or placeholder model. The app also now trusts the operating system's certificate store, resolving "unable to get local issuer certificate" failures for endpoints signed by corporate certificate authorities.

When the backend hub fails to start, the app now reports why and waits up to 15 seconds instead of 8. First launch after an install or update can take 8 to 13 seconds on Windows while the new binary is scanned by security software.

Additional fixes: session renames now persist after a relaunch without clearing other state like pinning; closed CLI sessions no longer appear as empty sidebar entries; arrow-key navigation in slash and @-mention menus now auto-scrolls; free models no longer show a redundant badge; and local models (llama.cpp, Ollama, LM Studio) that hit output-token limits now compact the conversation and retry once instead of failing outright.

Model catalog update

Cline added a new provider, ai&, an OpenAI-compatible endpoint serving open-weight models from Japan. The model catalog was refreshed to 6,386 models total. Default models changed for 19 providers, with 11 of them — including GitHub Copilot and Vertex — switching to Claude Opus 5.5. Users on those providers without a manually selected model will see a different default going forward.

What this means

This is an infrastructure and reliability release, not a new model launch. The Linux port widens Cline Desktop's addressable developer base significantly, since Linux is the default environment for much server-side and DevOps work. The plugin slash command fix and diagnostics export address real friction points for power users extending Cline's behavior. The default-model shift to Claude Opus 5.5 across 11 providers is notable mainly because it changes behavior silently for anyone who hasn't pinned a model — a detail worth checking in team deployments.