lemonade-sdk/lemonade v10.3.0 on GitHub

Headline

OmniRouter unifies the best backend engines to give you true omni-modal LLM chat. Make and edit images, generate speech, and more with natural language. Open AI-compliant.
The desktop app is now based on Tauri instead of Electron, so it is 10x smaller.
Easily update to specific llama.cpp versions, or enable auto-update.
The GUI has a light mode theme!
ROCm 7.2 stable release, 7.12 preview release, and TheRock nightly builds are all supported. Preview is now the default.

Breaking Changes

lemonade_server.service is now lemond.service.
The amd_igpu and amd_dgpu fields in system-info have been consolidated to amd_gpu. nvidia_dgpu renamed to nvidia_gpu for consistency.
ROCm channel default changed from nightly to preview.

Lemonade Server

Operating System	Downloads
Windows	lemonade.msi (Server + App)
Ubuntu 24.04+	Launchpad PPA (Server) · lemonade-app-10.3.0-x86_64.AppImage (Warning: Known Issues!)
Fedora 43	lemonade-server-10.3.0.x86_64.rpm (Server) · lemonade-app-10.3.0-x86_64.AppImage (Companion Desktop App)
macOS (beta)	Lemonade-10.3.0-Darwin.pkg (Server + App)

lemonade-server-minimal.msi is deprecated and will be removed in a future release.

Other platforms? See our Installation Options for Docker, Snap, Arch, Fedora, Debian, and more.

Embeddable Lemonade

Portable binaries for bundling into your own installer. Run lemond ./ as a subprocess.

Platform	Download
Ubuntu x64	lemonade-embeddable-10.3.0-ubuntu-x64.tar.gz
Windows x64	lemonade-embeddable-10.3.0-windows-x64.zip

What's Changed

Thanks @Geramy, @agreenbhm, @anditherobot, @bitgamma, @danielholanda, @fl0rianr, @github-actions, @ianbmacdonald, @jeremyfowers, @martinlroth, @michaelburch, @muriloime, @osimarr, @ramkrishna2910, @sawansri, @siavashhub, @sofiageo, @superm1 for your awesome contributions to this release!

Click to expand changelog

Fix broken debug build on Linux by @agreenbhm in #1593
[codex] add appear-builtin model aliases by @jeremyfowers in #1590
fix: skip ROCm download for gfx1033/1035/1036 by @martinlroth in #1596
Remove completely stale code.md doc by @jeremyfowers in #1609
Update llama.cpp to b8766 by @github-actions[bot] in #1619
feat: add light theme by @sofiageo in #1618
Fix Claude Code Guide by @sawansri in #1620
extend load timeout in CLI by @bitgamma in #1621
Disable realtime VAD for manual commit mode by @michaelburch in #1607
Remove stale link by @ramkrishna2910 in #1628
Migrate desktop app from Electron to Tauri by @jeremyfowers in #1611
Add qwen3.6-35b-a3b by @sawansri in #1650
Rename lemonade-server.service to lemond.service by @superm1 in #1651
Fix FLM version check on Linux to accept installed version >= minimum required by @Copilot in #1652
fix: make /load declarative/idempotent to prevent TOCTOU race by @ianbmacdonald in #1604
ROCm-Backend: added and tested docker image functionality with ROCm by @siavashhub in #1587
Drop AMD iGPU vs dGPU designations by @superm1 in #1648
fix: use wss:// for WebSocket connections when served over HTTPS (#1583) by @Geramy in #1666
Add a new configuration option to prevent installation of executables by @superm1 in #1668
Merge NOTICE.md into LICENSE by @superm1 in #1670
Add DESIGN.md and update the website by @jeremyfowers in #1660
Add support for ROCm channels (For llama.cpp and stable diffusion) by @superm1 in #1460
Upgrading FLM to 0.9.39 to add Gemma4 support by @ramkrishna2910 in #1644
Add a filter arg to the list command by @jeremyfowers in #1676
Refactor: Sync log levels to UI and implement client-side filtering by @anditherobot in #1655
Update the hot models for mid april 2026 by @jeremyfowers in #1694
Add recent Phoronix articles and FLM Linux guide to news page by @jeremyfowers in #1695
Theme updates by @sofiageo in #1673
Fix NVIDIA VRAM detection on Windows (#1681) by @jeremyfowers in #1682
fix: fail fast on insufficient disk space instead of retry loop by @ianbmacdonald in #1534
CLI pull command: accept full huggingface urls by @jeremyfowers in #1693
Enable system llama.cpp backend on Arch Linux by @osimarr in #1702
pull: remove the variants cap in cli and gui by @jeremyfowers in #1698
CLI: check the models list before sending pull endpoint requests by @jeremyfowers in #1691
Switch self-hosted runner targeting to capability-based labels by @jeremyfowers in #1697
macOS: use posix_spawn instead of fork+exec for child processes and patch HOME so its not null and doesn't throws a segfault in llama.cpp by @Geramy in #1708
Add Qwen3.6-27b by @sawansri in #1710
lemond: respond to SIGHUP by rescanning hardware and recipes by @Copilot in #1685
Prefer system backend by default generally by @superm1 in #1704
config: support "latest" and version tags for *_bin keys by @jeremyfowers in #1713
Add OmniRouter: multimodal tools + model collections for agentic workflows by @ramkrishna2910 in #1559
Add bonsai family of models by @sawansri in #1707
Update release notes downloads table by @jeremyfowers in #1719
Website touchups by @jeremyfowers in #1718
Remove no longer supported apps from Marketplace by @danielholanda in #1723
Split server_spec.md into many files by @jeremyfowers in #1721
docs(docker): document Web UI logs setup by exposing WebSocket port by @siavashhub in #1733
Consolidate sd ci jobs and remove stx-halo requirement by @jeremyfowers in #1732
Fix a return before closing file descriptor on linux npu utilization check. by @Geramy in #1735
fix(whisper): honor global_timeout instead of hard-coded 300s by @muriloime in #1737
Modify fallback logic for GGUF variant handling by @fl0rianr in #1727
Update llama.cpp to b8940 by @github-actions[bot] in #1743
ci(opensuse): unblock zypper install during transient libsystemd0 skew by @jeremyfowers in #1740
Fix realtime stop commit after VAD final by @jeremyfowers in #1739
chore: update light theme by @sofiageo in #1742
Refine collection image reply behavior by @jeremyfowers in #1726

New Contributors

@agreenbhm made their first contribution in #1593
@martinlroth made their first contribution in #1596
@michaelburch made their first contribution in #1607
@anditherobot made their first contribution in #1655
@osimarr made their first contribution in #1702
@muriloime made their first contribution in #1737
@fl0rianr made their first contribution in #1727

Full Changelog: v10.2.0...v10.3.0

Windows installers are signed. Free code signing provided by SignPath.io, certificate by SignPath Foundation. See our Code Signing Policy.