🚀 New Features

• Add one-click Kilo, Pi, Goose and OpenHands on the integration page to quickly start coding agents
• Enable "Launch at startup" by default for all users
• Render inline image attachments directly inside chat bubbles so images appear in-thread

🔧 Improvements & Fixes

• Prevent non-MTP models from being bricked by gating the global llamacpp-upstream MTP flag on a per-model capability at load time.
• 5+ stability and UI fixes

🙏 Contributors

Thanks to @Vect0rM and @danyurkin for their contributions to this release!

🚀 New Features

• Launch at startup — new toggle in Settings → General to start Atomic Chat automatically when your system boots (macOS, Windows, Linux)

🔧 Improvements & Fixes

• macOS recommended model now loads out of the box — the default llama.cpp engine is now vanilla upstream, so the Gemma vision model no longer crashes on load. TurboQuant stays available as an opt-in
• Readable model-load errors — failed loads now show the real reason instead of [object Object]
• Clearer Trusted Hosts setup — reaching the local API server over LAN now tells you exactly which address to add (use * to allow all)
• "Launch With" list — added Kilo Code, sorted by popularity. See the full list in the README

🙏 Contributors

Thanks to @Vect0rM, @yanalialiuk!

Atomic Chat v1.1.104

🚀 New Features

• Gemma 4 12B now runs in Atomic Chat — Google's latest multimodal model, working on Mac (MLX) and Windows/Linux (llama.cpp).
• Added Google's Gemma 4 QAT models support — near-original quality using up to ~72% less memory, with sizes from 2B up to 31B.

🔧 Improvements & Fixes

• Updated llama.cpp to the latest build, enabling faster Gemma 4 generation (MTP speculative decoding).
• Fixed a Windows bug where a fresh install couldn't save its settings, which could break first-run onboarding.
• Improved CUDA backend installation on Windows to fix cases where it failed to set up after updating.
• Future-proofed Windows GPU acceleration so it won't silently drop to CPU when the CUDA build version changes.
• Fixed the OpenClaw config being rejected so one-click Run works again without manual edits.

🙏 Contributors

Thanks to @Vect0rM, @Albert-Atomic for their contributions to this release!

🛠️ Fixes & Stability

• openclaw.json now supports JSON5 parsing for more flexible configuration
• Improved asset-aware download process and resolved download failures for llamacpp-upstream on Windows and Linux

🙏 Contributors

Thanks to @Vect0rM for their contributions to this release!

🚀 New Features

• Added a one-click OpenClaw agent launch

🔧 Improvements & Fixes

• Tool calling & MCP now work without reasoning enabled (#46).
• Windows CUDA-13 backend now resolves the correct CUDA minor version dynamically from the actual release assets (#43).
• Unsupported gemma4a multimodal projectors are now handled gracefully instead of crashing the whole llama-server with an opaque error (#44).

🙏 Contributors

Thanks to @GodsQuantum, @Vect0rM, @Albert-Atomic, @dtorey-d and @danyurkin for their contributions to this release!

🚀 New Features

• Added Hermes Agent, Codex CLI, Claude Code and Droid as one-click launch agents
• DFlash speculative decoding support on MLX-VLM for faster generation

🔧 Improvements & Fixes

• Bumped the Windows CUDA backend for llama.cpp from 13.1 to 13.3
• Artifacts now have a full-height resizable HTML preview with animations, an input-matched card, better code wrapping and fade-in for hub variants

🙏 Contributors

Thanks to @danyurkin, @Vect0rM, @dtorey-d, @Albert-Atomic for their contributions to this release!

🚀 New Features

• Integrations tab — launch the OpenCode and GitHub Copilot CLI agents in one click, backed by the local API server
• Artifacts — a live preview panel for HTML/CSS/JS code with copy, download and print buttons
• New MLX-VLM support — faster, leaner models:
  • EAGLE-3 speculative decoding for Gemma 4
  • MTP for Qwen 3.5/3.6 and DeepSeek V4
  • TurboQuant KV cache with RHT-correct fast paths for a smaller memory footprint

🔧 Improvements & Fixes

• Quit now properly terminates llamacpp-upstream child processes — no more stuck llama-server instances (#31)
• Redesigned the hub model card for clearer, better-organized model info
• The send button no longer stays disabled on the new-chat screen while a local model is still loading
• Added Hermes Desktop to the "Launch With" table in the docs
• 5+ more stability and UI fixes, including dark theme

🙏 Contributors

Thanks to @evil-doer, @Vect0rM, @Albert-Atomic, @dtorey-d, @yanalialiuk for their contributions to this release!

Atomic Chat v1.1.90

🚀 New

• Atomic Chat now runs on Linux as an AppImage — you just download and run it
• We added a curated catalog of hand-picked models and reworked search, so you can find the model you need right away

🔧 Improvements & Fixes

• Model downloads are now faster and smoother, the interface shows clear progress while they run
• We improved how the app detects your hardware and picks the right provider, with better support for Linux systems
• We fixed more than 10 smaller stability and interface issues

🙏 Contributors

Thanks to @Vect0rM and @dtorey-d for their contributions to this release!

Atomic Chat v1.1.83

🔧 Improvements

• Improved CUDA device enumeration and health checks to more reliably detect GPU availability and recover from driver issues
• Improved CUDA driver handling and GPU enumeration on Windows for more consistent detection across driver versions

🙏 Contributors

Thanks to @Vect0rM for their contributions to this release!

Atomic Chat v1.1.82

🔧 Improvements & Bug Fixes

• Refactored the local API server integration for more reliable connections to local backends. If you use custom local API configurations, please verify they still work — you may need to reconfigure them.
• Added a Local API Server tab in the sidebar for easier access, along with a one-click button to copy the server URL.
• Reworked Hugging Face authentication for a more streamlined token flow. You may need to re-enter your Hugging Face token in Settings after updating.
• Improved CUDA runtime handling in the llama.cpp upstream extension with more reliable cudart downloads.
• Refreshed assistant instruction handling for clearer and more predictable behavior.
• Updated model name constants to match the latest provider and model versions.

🙏 Contributors

Big thanks to @Vect0rM for code contributions to this release!