Repository navigation
Add llmman to Install > AI and Remove > AI - #438
ericcurtin wants to merge 4 commits into
Conversation
llmman runs the coding agents Omarchy already ships against a model on this machine: `llmman launch claude --model qwen3.8` fetches the tested upstream llama.cpp build for the GPU, loads the model and execs the agent. Its daemon speaks the Ollama, OpenAI and Anthropic APIs on one endpoint, models are plain OCI artifacts from Docker Hub or Hugging Face, and --provider points the same commands at a hosted model. It belongs next to LM Studio, Ollama and MLX as another way to run local models, and upstream ships an aarch64 Linux binary. The package is omarchy-pkgs' llmman-bin (omacom/omarchy-pkgs#428). The Install row dims on the command rather than the package, like Ollama's, because a cargo or curl install provides the same command; the Remove row keys on the package, which is the only thing omarchy-pkg-drop can act on. The remover stops the detached `llmman serve` daemon first (it has no stop command; pkill on its command line is upstream's own advice), drops the package, then deletes the store, config and the two files llmman writes into other agents' homes under its own name. The agents' own state beside them stays, and the tests pin that boundary. Same change as omacom#11722, carried here because this menu has diverged from upstream's.
|
Independent review (AI-assisted, Claude). Read the remover, the menu rows and the test; ran the QA set in an Ubuntu 24.04 arm64 container. Dependency status Both prerequisites are still open: omacom/omarchy-pkgs#428 ( Code
Local verification (Ubuntu 24.04 arm64 container)
|
|
@maralcbr @ryanrhughes PTAL when you get a chance. It needs omacom/omarchy-pkgs#428. Thank you! |
Signed-off-by: Eric Curtin <eric.curtin@docker.com>
|
Thank you @laihenyi, the HOME check now runs first in 5ad6c35 (same fix on omacom#11722). I left the pkill pattern as is. |
Signed-off-by: Eric Curtin <eric.curtin@docker.com>
Signed-off-by: Eric Curtin <eric.curtin@docker.com>
|
@dhh @joshuaswarren PTAL when you get a chance. It needs omacom/omarchy-pkgs#428. If useful, please also try https://github.com/llmmanorg/llmman. Thank you! |
Adds llmman beside MLX, LM Studio and Ollama under Install > AI, with a matching Remove > AI entry, following the shape of the Ollama pair. Same change as omacom#11722, carried here because this menu and test file have diverged from upstream's; if you'd rather pick it up through the next sync from omarchy instead, feel free to close this one.
llmman is the piece that points the coding agents Omarchy already ships at a local model:
llmman launch claude --model qwen3.8fetches the tested upstreamllama.cppbuild for the machine, loads the model and execs Claude Code against it (same foropencode,codex,agy,aider, ...).llmman serveis one endpoint on127.0.0.1:17434speaking the Ollama, OpenAI and Anthropic APIs, so anything already talking to Ollama works unchanged; models are plain OCI artifacts pulled straight from Docker Hub or Hugging Face;--providerruns the same agent against a hosted model instead. Upstream publishes anaarch64-unknown-linux-gnubinary, and the PKGBUILD carriesarch=('x86_64' 'aarch64').Depends on the package: omacom/omarchy-pkgs#428 adds
llmman-bin.What's here
install.ai.llmman-omarchy-install-app llmman llmman-bin. Dims onomarchy-cmd-present llmmanrather than the package, like the Ollama row, since a cargo or curl install provides the same command.remove.ai.llmman- shown onomarchy-pkg-present llmman-bin, the only thingomarchy-pkg-dropcan act on.bin/omarchy-remove-ai-llmman- underset -euo pipefail. Stops the detachedllmman servedaemon first (it has no stop command;pkill -f 'llmman serve'is upstream's own advice and matches nothing but the daemon), drops the package, then deletes~/.local/share/llmman(model store, downloaded llama.cpp releases, prompt history),~/.config/llmman(aliases, provider keys, trust policy), and the two files llmman writes into other agents' homes under its own name:~/.codex/llmman.config.tomland~/.gemini/llmman/. Everything else in~/.codex,~/.geminiand~/.hermesstays.manual/17-ai.md- Local LLMs now names three options and says what llmman is for.test/shell.d/remove-ai-test.sh- adds apkillstub (so the suite never reaches a developer's own daemon) and pins what the remover deletes, what it keeps, that it dropsllmman-binand pkills the daemon, that it refuses withoutHOME, and that the Remove row keys on the package.Icon
A Nerd Font glyph (
nf-md-server) rather than a new mark in the Omarchy icon font: upstream's only artwork is a raster manatee with no monochrome SVG.Validation
In
archlinux:base-devel:remove-ai-test.sh(17 ok, including the 6 new),menu-test.sh(121 ok),bin-style-test.sh,desktop-entry-launch-test.shand./test/cli(119 ok, so the new command's metadata lints) all pass.