{
"markdown": "\n\n# Local AI for Omarchy\n\n**Your Framework Desktop. Your models. A terminal away.**\n\nA local coding workspace for a freshly installed [Omarchy](https://omarchy.org/)\nsystem on the **Framework Desktop, Ryzen AI Max+ 395, 128 GiB**. Open a compact\nkeyboard-driven menu, keep an everyday model warm, and give each project a\npersistent [Herdr](https://herdr.dev/) workspace. Leave your agent, build, and\nlogs in place; return to them from your desktop or over SSH.\n\n[Get running](#make-yourself-at-home) · [Meet the models](#a-small-model-team) ·\n[Herdr workflows](docs/herdr.md) · [Workstation guide](docs/omarchy.md) · [Full reference](docs/reference.md)\n\n| At your keyboard | On your machine | Under your control |\n|---|---|---|\n| Herdr project workspaces and a pinned OMP agent | One local model at a time, accelerated by RADV / Vulkan | Plain configuration, reviewed downloads, a systemd user service |\n\nThe experience follows Omarchy's terminal, theme, and personal-configuration\nconventions. The Strix Halo memory and inference settings remain the same.\n[See how the runtime fits together.](docs/reference.md#llamacpp-router-and-per-model-presets)\n\n## Make yourself at home\n\nOpen an Omarchy terminal with Super + Return and update the system:\n\n```bash\nomarchy update\n```\n\nReboot if requested, then clone and inspect this repository. Keep the checkout:\nthe installed command points back to it. Omarchy's updater handles its\nsnapshots and migrations. [Update guide](https://omarchy.org/manual/updates/).\n\n```bash\nmkdir -p ~/github\ngit clone https://github.com/johnymontana/local-ai-setup.git ~/github/local-ai-setup\ncd ~/github/local-ai-setup\ngit rev-parse HEAD\nless install.sh\nless setup-qwen38-pi.sh\n./install.sh\n```\n\nRun as your ordinary desktop user; the installer asks for `sudo` only for\nsystem changes. It checks for pending updates, installs the runtime packages,\neveryday model, pinned agent and Herdr, enables the router's user service, and\nadds **Local AI**, **Local AI Logs**, and **Local AI Workspaces** to app search.\nIf it requests a reboot, reboot and run `./install.sh`\nagain. Downloads resume. For repeatable deployments, use a\n[reviewed commit](docs/reference.md#reproducible-downloads-and-installs).\n\n> [!TIP]\n> **Start with Everyday.** The installer sets up the baseline; add Coder and\n> Senior when a task calls for them. Keep this checkout in its permanent home.\n\n> [!NOTE]\n> **If the package step stops.** Arch reorganizes the llama.cpp and ggml\n> packages from time to time — the mandatory CPU backend moved out of\n> `ggml-cpu` and into `ggml`, and the two names now conflict. The installer\n> resolves the current names against your channel, so update this checkout\n> first if you see `unresolvable package conflicts detected`. When a superseded\n> `ggml-cpu` is still installed from an earlier setup, run `omarchy update`\n> once so the native updater performs the replacement, then run `./install.sh`\n> again. [Other install failures](docs/reference.md#verification-and-troubleshooting).\n\nAfter installation, open a new terminal:\n\n```bash\nlocal-ai status\nlocal-ai smoke everyday\ncd ~/github/your-project\nlocal-ai workspace\n```\n\nUse `local-ai` to open the menu, or search for **Local AI** in Omarchy's app\nlauncher. `status` reads state; `smoke` intentionally loads a model and generates\na response. [The workstation guide](docs/omarchy.md) covers setup, desktop\nintegration, updates, backups, and recovery.\n\n`workspace` opens the current project with a lead agent and dedicated terminal\nroles. Reopening reuses the workspace. `local-ai-agent` remains available for\na single terminal session. Set `HERDR_ENABLED=0` when running `./install.sh` to\nskip Herdr, or add it later with `local-ai herdr`.\n\n\n\n*The real plain-terminal menu in an isolated pre-install demo, rendered with an\nEverforest-inspired palette. [Capture details and text version](docs/assets/README.md).\nYour live menu follows your selected terminal theme.*\n\n## A place for work to stay\n\n\n\n*An illustrated workspace map in the documentation's Everforest palette.\nHerdr itself follows your active Omarchy terminal colors.*\n\n```bash\nlocal-ai workspace open ~/github/your-project\nlocal-ai workspace open ~/github/golf-game --profile golf\nlocal-ai workspace list\nlocal-ai workspace attach ~/github/your-project\n```\n\n**Coding** makes room for Lead, Shell, Tests, Build, Logs, and Status. **Golf**\nadds Physics, Course, Rendering, and Audio roles, ready for explicit tasks.\n**Operations** opens the local router journal, status, and a shell without\nstarting a coding agent. Only Lead starts automatically in a new coding or golf\nworkspace; extra panes do not mean extra resident models.\n\nThe [Herdr guide](docs/herdr.md) turns those roles into working build/test\ncommands, serialized specialist delegation, a golf development loop, and\npersistent SSH access. Menu shortcuts **w** and **o** open project and\noperations workspaces; **h** installs the pinned Herdr layer.\n\n## A small model team\n\n\n\n| Model | Best place to start | Locked download |\n|---|---|---:|\n| **Everyday** · Qwen 3.8 27B | Implementation, tests, shell work, and vision | ~18.5 GiB |\n| **Coder** · Qwen3-Coder-Next | Repository exploration, tools, and debugging | ~45.1 GiB |\n| **Senior** · Qwen3.5-122B-A10B | Planning, difficult bugs, and review | ~69.5 GiB |\n\nThe installer starts with Everyday. Allow **24 GiB free** for its default Q4\nartifacts and the 5 GiB reserve. All three tiers need about **138 GiB free**;\nusing Everyday Q8 raises that to about **149 GiB**. Every artifact is pinned by\nrevision, size, and SHA-256 in [models.lock](models.lock). Senior uses an optional\ncommunity Unsloth quant.\n\nAdd a specialist when needed:\n\n```bash\nlocal-ai model coder\nlocal-ai plan\nlocal-ai apply\nlocal-ai smoke coder\ncd ~/github/your-project\nomp-coder\n```\n\nUse `senior` in the same workflow for the larger reviewer. `omp-everyday`,\n`omp-coder`, and `omp-senior` select a phase explicitly; launchers exist for\ncomplete installed tiers. The default sticky routing keeps one model warm and\none request in flight, leaving room for the desktop. Omarchy's own `omp` and\n`pi` commands stay available; use this project's launchers for its pinned stack.\n\n## Same hardware, considered defaults\n\nKeep the BIOS UMA framebuffer at its small/default **512 MiB** and **IOMMU\nenabled**. Start with stock GTT. Vulkan offload, per-model context, MTP,\nload modes, and the single-resident-model limit retain the existing Framework\nDesktop tuning. The optional **115 GiB GTT** setting is for measured needs;\nSenior can use mixed CPU/GPU loading at the stock limit.\n\n> [!NOTE]\n> These are configuration defaults. Run the\n> [on-machine checks](docs/omarchy.md#verify-on-the-workstation) after installation\n> and driver updates to verify Vulkan offload and useful desktop headroom.\n\n## Keep it yours\n\nConfiguration lives in `~/.config/local-ai`, models in `~/llm/models`, and the\nrouter runs as your user. The menu follows your terminal's colors and uses\nOmarchy's palette when available. Personal settings and desktop integration\nfollow [Omarchy's dotfile conventions](https://omarchy.org/manual/dotfiles/).\n\nInference stays local. Coding agents can still read files, run commands, and\nuse the network with your account's permissions; read the\n[security boundaries](docs/reference.md#security-boundaries) before using\nuntrusted repositories.\n\n| Next stop | What you’ll find |\n|---|---|\n| [Workstation guide](docs/omarchy.md) | The install path, daily workflow, updates, and recovery |\n| [Herdr workflows](docs/herdr.md) | Persistent projects, golf roles, explicit builds, delegation, and SSH |\n| [Full reference](docs/reference.md) | Model routing, tuning, commands, remote access, and security |\n| [Contributor checks](docs/reference.md#contributor-checks) | Portable tests and opt-in hardware verification |\n| [Performance implementation record](docs/performance-dx-plan.md) | Design rationale and the layers of evidence |\n\n*Visuals take their cues from [Omarchy’s Everforest theme](https://omarchy.org/manual/themes/).\nThis is an independent local-AI setup project.*\n",
"bytes": 8810,
"sha": "80e6661f5ddf4f8d3805786e33d5dd406e82f1211c4c5768cf99bc7a47ab1667",
"repo_slug": "johnymontana/local-ai-setup",
"fonte": "repo",
"truncated": false,
"api": "https://agentalog.com/api/listings/okf_johnymontana_local_ai_setup_openwiki_ind_d6f594c8/readme"
}