Skip to content

ContextControl v0.6.2 - Linux, cloud inference and model controls

Latest

Choose a tag to compare

@VulkanVX VulkanVX released this 17 Sep 19:14

ContextControl v0.6.2

ContextControl now has a Linux desktop package and can send inference to another computer from the same Windows/Linux app. Model allocation is directly configurable, with separate requested settings, hardware estimates and observed runtime allocation.

  • Linux desktop: the same project, context workflow, model catalog and chat codebase. The archive includes .NET and the native exporter; PowerShell 7 supports GO/Apply. Ollama installs into the user directory from verified official archives and runs as a user service.
  • Cloud Machines: configure inference computers, reuse SSH identities, import/export connection presets, and switch between local and server models in chat. A compact green/grey indicator shows reachability and connection latency. Cloud toggles preserve the model list and no longer flash the global progress panel.
  • Model controls: edit context, GPU layers, CPU threads, prompt batch size, output limit, temperature and top-p per model and connection. Adaptive allocation preserves manual values and estimates fit using the selected machine's available memory, context/cache, offload split, workspace and reserves. Ollama servers receive native allocation options through the private connection; other compatible servers retain their allocation controls. Unknown remote hardware is never replaced with laptop specifications.
  • Clear model facts: installed metadata exposes parameter count, quantization, transformer blocks and context limits. The 1,980-entry catalog adds numeric parameter sorting, compact release dates, current model families and quantization download options. CPU suitability uses runtime support and RAM requirements rather than a static CPU incompatibility label. Dedicated GPU capacity filters keep replacement models visible while another model is loaded.
  • Android-style emoji: bundled Google Noto Color Emoji artwork on Windows and Linux, preserving joined sequences, skin tones, flags and keycaps in chat, markdown, wrapping and truncation.
  • Chat and workflow polish: retains the current action, conversation-memory and project-context improvements, and fixes hidden Game Lab previews keeping the app alive after the main window closes. PowerShell rule fingerprints remain deterministic across processes.

Install

  • Windows: download ContextControl-win-x64-Setup.exe. Existing settings and chats are preserved during updates.
  • Linux x64: download and extract ContextControl-linux-x64.tar.gz, then run start-contextcontrol.sh. install-desktop.sh adds the menu entry.
  • SHA-256 files accompany both packages. Model weights are downloaded separately inside the app.

See the Linux and cloud setup guide, including Model controls and connecting another PC. Linux's embedded WebView2 browser and browser-dependent previews remain unavailable; external browsers are supported. Memory fit is an estimate, not a speed or model-quality guarantee. Ollama's allocation API does not report observed thread, batch or GPU-layer counts.

Validation

Validated on Windows and Ubuntu with a Ryzen 9 9950X and Radeon RX 7900 XTX. Checks cover the full UI, resource planning, settings persistence, native/compatible chat, SSH routing, model controls and emoji rendering. Live SSH inference verified CPU-only and full-GPU requests, adaptive context using server hardware, and the actual chat toggle. A complete 27B model download and subsequent chat also passed. The native Windows process exits correctly with a hidden preview window.

Built and packaged locally from the release source. No GitHub Actions run is claimed.