Osmantic/ODS

ODS installs and wires together everything you need to run AI locally, so you do not have to assemble Ollama, Open WebUI, n8n, ComfyUI, and privacy tools by hand:

• Local model inference — run open models on your own hardware
• ChatGPT-style web UI — talk to your models from any browser
• Control dashboard — manage models, services, setup, GPU status, and extensions from one place
• Voice, agents, and workflows — build automations that can listen, speak, call tools, and get work done
• RAG and search — connect local documents, private search, and retrieval workflows
• Image generation — run local image tools without sending prompts to a hosted API
• Privacy and ops — keep service auth, secrets, observability, and diagnostics in one local stack

Status: V3 pre-release. The installers below install the current main

branch, which receives fixes continuously and is not a signed release; the

verified installer preview stays

separate until a signed release passes end-to-end testing.

Existing installations pick up code fixes by re-running the installer

(how); ods update refreshes

container images only. v3.0.0 is the latest published source release; check the

security advisories before

pinning it. Known limits and validation

status: V3 release notes,

Release Validation,

Release Channels and

Installer Trust.

The repository root holds this README, the installers, the security policy and CI workflows; ods/ is the product: services, installer phases, compose files, dashboard, CLI, tests and operator docs.

Get Started

Choose your system, copy the block, run it in a normal terminal. ODS installs the stack, picks a model for your hardware, starts the services, and gives you the local web UI.

Linux or macOS

curl -fsSL https://install.osmantic.com/ods.sh | bash

Windows PowerShell — guided Ubuntu/WSL2 setup with Pixel/Portal

& {
    $ErrorActionPreference = 'Stop'
    $ProgressPreference = 'SilentlyContinue'
    $odsSrc = Join-Path $env:TEMP ('ods-install-' + [guid]::NewGuid().ToString('N'))
    $odsZip = Join-Path $odsSrc 'ods-main.zip'
    New-Item -ItemType Directory -Path $odsSrc | Out-Null
    Invoke-WebRequest -UseBasicParsing 'https://github.com/Osmantic/ODS/archive/refs/heads/main.zip' -OutFile $odsZip
    Expand-Archive -LiteralPath $odsZip -DestinationPath $odsSrc
    $odsEntry = Join-Path $odsSrc 'ODS-main\install.ps1'
    if (-not (Test-Path -LiteralPath $odsEntry -PathType Leaf)) { throw 'The downloaded archive does not contain the ODS installer.' }
    Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass
    & $odsEntry
}

Linux and macOS: Docker must be installed and running.

Windows: open a normal PowerShell window (not "Run as administrator"), paste the block, and answer the prompts. Nothing else needs to be installed first. The installer:

1. Checks free disk space (40 GB) and that hardware virtualization is on. 2. Offers to enable WSL2 and install Docker Desktop with winget. Windows asks for administrator permission, then one restart; setup continues by itself after you sign in again. 3. Offers to download Ubuntu 24.04 and asks you, in PowerShell, for a new Ubuntu username and password. 4. Starts Docker Desktop and checks that it is connected to Ubuntu. If not, it shows the one setting to turn on in Docker Desktop and continues as soon as it works. 5. Installs ODS inside Ubuntu with --pixel --no-hermes. A rerun leaves out --no-hermes, so a Hermes you added from Extensions stays on. When Ubuntu asks for your [sudo] password, type the Ubuntu password; nothing appears while you type. 6. Verifies Pixel and Portal, then opens Portal in your browser and creates the ODS app on your desktop and in the Start menu: Portal in its own Microsoft Edge window, with the ODS icon. Without Edge it adds an ODS Portal web link to your desktop instead.

Each step asks before changing anything and stops with instructions if it cannot finish; rerun the same command after fixing it. There is no fallback to Hermes or the native Windows installer. On NVIDIA machines, update the Windows driver to 570 or newer first. To use an existing distribution, add -Distro (names from wsl -l -v).

Existing native Windows installations are not automatically migrated or deleted; see Windows Quickstart before switching.

If another device runs your ODS model gateway and this Windows PC only needs image generation, use the standalone ComfyUI installer. It keeps its Docker project and data separate from a full ODS installation.

The hosted Linux/macOS endpoint proxies the current bootstrap from repository main. Reviewed merges reach it automatically after edge-cache refresh. ODS_REF selects a compatible repository checkout. See Installer Trust to inspect the script or install an audited commit manually; no release has qualified for the verified channel yet.

Windows users should not run the curl ... | bash command from PowerShell. The PowerShell block above downloads the public ODS source ZIP and delegates installation to Ubuntu/WSL2. For more detail, see the Windows Quickstart.

After the installer completes successfully, Portal opens at http://localhost:3001/pixel (the Windows installer opens it for you and prints the exact URL). http://localhost:3000 is Open WebUI, a separate interface. Verify that Portal is available and send a message; a loaded dashboard alone does not prove Pixel is ready. If installation fails or Portal is degraded, follow the Windows Quickstart checks before proceeding.

WSL GPU access must be checked separately. NVIDIA needs a supported Windows driver and GPU access inside WSL/Docker. On AMD, Windows setup runs llama.cpp's llama-server.exe (Vulkan) through an ODS task bound to the selected WSL installation. Once that ownership is verified, Dashboard Models supports compatible GGUF downloads (including Hugging Face), activation, context changes, and unload/resume. A model server that the installation does not manage remains externally managed. Installations whose ODS task ran Lemonade Server move to llama.cpp when the installer runs again; see the Windows Quickstart, the WSL2 GPU guide and AMD GPUs now run on llama.cpp.

For Linux, macOS, or the recommended Windows/WSL installation, uninstall from the matching Linux/macOS terminal (open Ubuntu on Windows):

cd ~/ods
./ods-uninstall.sh --force

For a native Windows installation only:

$installDir = "$env:USERPROFILE\ods"
cd $installDir
.\ods.ps1 uninstall --force

Windows recovery note: if the runtime folder is partial and .\ods.ps1 is missing, run the same command from a source checkout as .\ods\installers\windows\ods.ps1 uninstall --force. It verifies the containers' Compose installation directory before removing resources. A shared ods project name does not authorize removing another Windows or WSL installation. Unattached volumes with no verifiable owner are preserved, with an error naming the resource; --force does not bypass this check.

API endpoint: Linux Docker installs expose llama-server on http://localhost:11434 by default (OLLAMA_PORT) while containers use llama-server:8080. macOS native Metal and the Windows AMD llama-server.exe use http://localhost:8080 unless overridden; the Windows server requires its API key. Open WebUI stays on http://localhost:3000.

No GPU? ODS also runs in cloud mode — same full stack, powered by OpenAI/Anthropic/Together APIs instead of local inference:

> ./install.sh --cloud

Port conflicts? Every port is configurable via environment variables. See .env.example for the full list, or override at install time:

> WEBUI_PORT=9090 ./install.sh

New here? Read the Friendly Guide or listen to the audio version — a complete walkthrough of what ODS is, how it works, and how to make it your own. No technical background needed.

---

At A Glance

| Question | Answer | |----------|--------| | What is it? | A local AI server stack for your own hardware, with a one-command Linux/macOS installer and a PowerShell installer for Windows. | | Who is it for? | People who want private AI at home, in a lab, or on a workstation without hand-wiring a dozen services. | | What do I get? | Local inference, Open WebUI chat, a control dashboard, voice, agents, workflows, RAG, search, image generation, privacy tools, observability, and developer tools. | | What does it run on? | Linux, Windows with WSL2/Docker Desktop, and macOS Apple Silicon. | | Is cloud required? | No. Local mode is the default; cloud and hybrid API modes are optional. |

| If you know... | ODS adds... | |----------------|----------------------| | Ollama / llama.cpp | The surrounding server stack: chat, dashboard, voice, RAG, workflows, agents, privacy, and service management. | | Open WebUI | A full installer and control plane around Open WebUI, plus pre-wired local services. | | AnythingLLM | Broader local AI appliance behavior beyond RAG: inference, chat, voice, workflows, image generation, and ops. | | n8n self-hosted AI starter kits | Workflow automation as one part of a larger private AI server. |

---

Current Platform Support

> | Platform | Status |

|----------|--------|

| Linux (NVIDIA + AMD Strix Halo) | Supported — see the hardware and distro limits in the support matrix |

| Linux + Intel Arc (SYCL) | Experimental / Tier C — validation is hardware-specific |

| Windows (NVIDIA + AMD) | Supported — install and run today |

| macOS (Apple Silicon) | Supported — install and run today |

> Linux distros: CI smoke-tests package-manager detection and installer prerequisites on Ubuntu 26.04/24.04/22.04, Debian 12, Linux Mint 21.3, Fedora 41, Rocky Linux 9, Arch Linux, Manjaro, CachyOS and openSUSE Tumbleweed. Broader install coverage runs in the maintainers' distro lab and hardware fleet described in the Validation Matrix. On openSUSE the installer installs Docker from the distribution's own packages. Open an issue if your distro doesn't work.

> Release validation: Operational changes run through a release-grade gate

that covers zero-prereq bootstrap, clean installs, product behavior,

full-model capabilities, lifecycle recovery, and User Green. See

Release Validation and the

Validation Matrix.

> Windows: Requires Docker Desktop with WSL2 backend. NVIDIA GPUs use Docker GPU passthrough; AMD Strix Halo runs through the platform-specific accelerated path documented in the Windows installer and support matrix.

> macOS: Requires Apple Silicon (M1+) and Docker Desktop. llama-server uses native Metal acceleration; Portal's gateway and managed host helpers also run natively. The UI, ingress, sandbox and supporting services run in Docker. See the macOS Quickstart.

> See the Support Matrix for supported

platform claims and the Validation Matrix

for the layered test surface used to test those claims.

---

Why ODS?

A handful of companies control the vast majority of global AI traffic — and with it, your data, your costs, and your uptime. Every query you send to a centralized provider is business intelligence you don’t own, running on infrastructure you don’t control, priced on terms you can’t negotiate.

If AI is becoming critical infrastructure, it shouldn’t be rented. Self-hosting local AI should be a sovereign human right, not a career choice.

Because running your own AI shouldn't require a CS degree and a weekend of debugging CUDA drivers. Right now, setting up local AI means stitching together a dozen projects, writing Docker configs from scratch, and praying everything talks to each other. Most people give up and go back to paying OpenAI.

We built ODS so you don't have to.

• One command — detects your GPU, picks the right model, generates credentials, launches everything
• Chat early — bootstrap mode starts a small model as soon as the core services are up, while your full model downloads in the background
• Pre-wired services — the default Core install gives you local inference, the dashboard and a chat UI (the Portal agent on supported hosts, Open WebUI elsewhere); add voice, workflows, search, RAG, image generation, observability and developer tools from the installer menu or the Extensions page, already wired to each other
• Fully moddable — every service is an extension. Drop in a folder, run ods enable, done

git clone https://github.com/Osmantic/ODS.git
cd ODS/ods
./install.sh

$ProgressPreference = "SilentlyContinue"
$odsSrc = Join-Path $env:TEMP ("ods-install-" + [guid]::NewGuid().ToString("N"))
$odsZip = Join-Path $odsSrc "ods-main.zip"
New-Item -ItemType Directory -Path $odsSrc | Out-Null
Invoke-WebRequest "https://github.com/Osmantic/ODS/archive/refs/heads/main.zip" -OutFile $odsZip
Expand-Archive -LiteralPath $odsZip -DestinationPath $odsSrc -Force
cd (Get-ChildItem -LiteralPath $odsSrc -Directory | Select-Object -First 1).FullName
Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass
.\install.ps1

git clone https://github.com/Osmantic/ODS.git ~/src/ODS
cd ~/src/ODS/ods
./install.sh

---

What's In The Box

Chat & Inference

• Open WebUI — full-featured chat interface with conversation history, web search, document upload, and 30+ languages
• llama-server — high-performance LLM inference with continuous batching, auto-selected for your GPU; Linux Docker host API defaults to localhost:11434, native macOS/Windows paths use localhost:8080, and container API runs on 8080
• LiteLLM — API gateway supporting local/cloud/hybrid modes
• TEI Embeddings — text embedding service for RAG and search workflows

Voice

• Whisper — speech-to-text
• Kokoro — text-to-speech

Agents & Automation

• Portal — bundled core conversational assistant on Apple Silicon macOS and qualified Ubuntu 24.04/26.04 or Debian 12 systemd hosts, including qualified WSL2 installations through the Linux installer. No private repository access or separate license flag is required; available in the Dashboard and through a compatible Open WebUI model route. The native PowerShell installer does not install the Portal host runtime.
• Hermes Agent — independent general-purpose agent, available alongside Portal; includes memory, skills, and a proxy with optional owner-card gating; direct access by default
• n8n — workflow automation with 400+ integrations (Slack, email, databases, APIs)
• APE — Agent Policy Engine for auditing and governing autonomous tool calls
• OpenCode — browser-based AI coding assistant wired to the local stack
• Memory Shepherd — host/systemd helper for agent memory lifecycle management

Knowledge & Search

• Qdrant — vector database for retrieval-augmented generation (RAG)
• SearXNG — self-hosted web search (no tracking)
• Perplexica — deep research engine
• Brave Search — optional paid Brave Search API integration

Creative

• ComfyUI — node-based image generation

Privacy & Ops

• Privacy Shield — PII scrubbing proxy for API calls
• Dashboard — real-time GPU metrics, service health, model management
• Dashboard API — service health, setup, status, metrics, and management API behind the dashboard
• Token Spy — token usage monitor for local and proxied LLM traffic
• Langfuse — optional LLM observability and tracing

Hardware Auto-Detection

The installer detects your GPU and first assigns a deterministic hardware tier. Linux and macOS then run the versioned catalog selector (ods/scripts/select-model.py), while Windows uses the PowerShell catalog selector in ods/installers/windows/lib/tier-map.ps1; both read ods/config/model-library.json to choose the best installable GGUF for the detected memory envelope. The final choice is written to .env as LLM_MODEL, GGUF_FILE, MAX_CONTEXT, and MODEL_RECOMMENDATION_*.

MODEL_PROFILE=qwen is the default catalog profile. MODEL_PROFILE=gemma4 forces Gemma 4 where available, and MODEL_PROFILE=auto uses Gemma 4 on NVIDIA, AMD Strix Halo, Apple Silicon and Intel Arc tiers. Override the model family with MODEL_PROFILE=gemma4 ./install.sh. You can switch models at any time after installing (see Switching Models).

The tables below are what the catalog selector picks today for common hardware with the default profile and a 64K context floor (python ods/scripts/select-model.py --profile qwen --installable-only --min-context 65536 …). Exact picks can differ with detected VRAM/RAM, host architecture and existing downloads. Measure throughput on your own machine after the first launch.

NVIDIA

| Memory | Default pick | Context | Example hardware | |--------|--------------|---------|------------------| | 8–16 GB VRAM | Qwen3.5 9B (Q4_K_M) | 64K | RTX 4060, RTX 3060 12GB, RTX 4070-class | | 24–32 GB VRAM | Qwen3.5 27B (Q4_K_M) | 64K | RTX 4090, RTX 5090 | | 48 GB VRAM | Qwen3.6 35B-A3B (UD-Q4_K_M) | 128K | RTX 6000 Ada, L40S | | 96 GB+ VRAM (x86_64) | Qwen3 Coder Next (Q4_K_M) | 128K | RTX PRO 6000, multi-GPU | | 128 GB unified (arm64) | Qwen3.6 35B-A3B (UD-Q4_K_M) | 128K | DGX Spark / GB10 |

AMD Strix Halo (Unified Memory)

| Memory | Default pick | Context | Hardware | |--------|--------------|---------|----------| | 64–128 GB unified | Qwen3.6 35B-A3B (UD-Q4_K_M) | 128K | Ryzen AI MAX+ 395 |

The selector routes unified-memory hosts away from Qwen3 Coder Next because of documented correctness issues on those backends.

Apple Silicon (Unified Memory, Metal)

| Memory | Default pick | Context | Example hardware | |--------|--------------|---------|------------------| | 8 GB | NVIDIA Nemotron3 Nano 4B (Q4_K_M) | 64K | M1/M2 base | | 16–36 GB | Qwen3.5 9B (Q4_K_M) | 64K | M4 Mac mini, M3 Pro | | 48 GB+ | Qwen3.6 35B-A3B (UD-Q4_K_M) | 128K | M4 Pro/Max, Mac Studio |

CPU only

| System RAM | Default pick | Context | |------------|--------------|---------| | 8–12 GB | Qwen3.5 2B (Q4_K_M) | 64K | | 16–20 GB | Qwen3.5 4B (Q4_K_M) | 64K | | 24 GB+ | Qwen3.5 9B (Q4_K_M) | 64K |

Intel Arc (Linux, SYCL)

Intel Arc is experimental (Tier C). Hardware detection does not identify Arc GPUs yet, so pass the tier yourself: ./install.sh --tier ARC_LITE for 6–8 GB cards or --tier ARC for 16 GB cards. See the Intel Arc guide.

---

Bootstrap Mode

ODS uses bootstrap mode by default so you can start before a large model finishes downloading:

1. The installer downloads Qwen3.5 2B (about 1.3 GB) and starts it. 2. You can chat as soon as the core services are up. 3. The full model for your hardware downloads in the background. 4. When the download is verified, ODS restarts the model server on the full model. Requests in flight during that restart need a retry.

The bootstrap model starts with a 64K context window so agents work during the first session. After the swap, ODS restores the full model's context target.

Skip bootstrap: ./install.sh --no-bootstrap

---

Switching Models

The installer picks a model for your hardware, but you can switch anytime. The recommended way is Dashboard → Models: it lists models that fit your hardware, downloads them with checksum verification, and loads them. If a model fails to start, ODS restores the previo