You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Release channel: the install commands on this page fetch development main, which is not signed. A signed-source path is staged in Verified Install Preview; it is not active until the first eligible immutable release is published, and historical v3.0.0 is not eligible.
One command to a running local AI stack. The installer detects your hardware,
chooses a model, writes the config, starts the services, and leaves you with a
chat UI plus the ods management command.
NVIDIA Container Toolkit for NVIDIA GPUs, the AMD GPU render node
(/dev/dri; also /dev/kfd for the optional ROCm image) for AMD, or Intel
compute runtime for Arc
40 GB+ free disk space for models and container images
macOS:
Apple Silicon Mac
Docker Desktop running
16 GB+ unified memory recommended
20 GB+ free disk space
Windows:
Windows 10/11
Docker Desktop with WSL2 backend enabled and running
NVIDIA GPU or AMD Strix Halo recommended
A normal user PowerShell session. Do not run the installer as Administrator
unless you deliberately want admin-owned files under your user profile.
The hosted 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.
Do not run this command from Windows PowerShell; use the Windows installer below.
The Windows command guides Ubuntu/WSL2 preparation and requires Pixel, with no Hermes fallback. It installs missing WSL, Docker Desktop and Ubuntu after asking, continues by itself after the one restart, and opens Portal when done. See Windows Quickstart.
Useful install flags:
Linux/macOS
Windows
Purpose
--all
-All
Enable the recommended full stack
--voice
-Voice
Enable Whisper STT and Kokoro TTS
--workflows
-Workflows
Enable n8n workflows
--rag
-Rag
Enable Qdrant and embeddings
--pixel
—
Require Pixel on a qualified Linux/systemd host
--no-pixel
—
Disable Pixel and use the configured fallback
--no-hermes
-NoHermes
Disable the Hermes fallback agent
--no-bootstrap
-NoBootstrap
Wait for the full model instead of fast-start
--tier 3
-Tier 3
Force a hardware/model tier
Uninstall
Linux/macOS, or inside Ubuntu for Windows/WSL:
cd~/ods
./ods-uninstall.sh --force
Legacy native Windows installations only:
$installDir="$env:USERPROFILE\ods"
cd $installDir
.\ods.ps1 uninstall --force
Use --keep-data or --keep-models to preserve local state. --keep-data
keeps only the data folder inside the install directory. It still deletes
.env (your settings and generated secrets) and config/, and on Linux and
macOS the backups in ~/.ods. If you plan to reinstall over the kept data,
copy those somewhere safe first and put .env back before running the
installer; without it, the installer generates new secrets.
If the Windows
runtime folder is partial and .\ods.ps1 is missing, run
.\ods\installers\windows\ods.ps1 uninstall --force from a source checkout.
What Happens First
Bootstrap mode is enabled by default when your selected full model is large.
ODS downloads a small model first so you can start chatting quickly,
then downloads and hot-swaps the full model in the background.
Pixel is installed automatically on qualified Ubuntu 24.04/26.04 or Debian 12
PID1-systemd hosts. The Windows entry point explicitly requires Pixel in WSL2 and stops on failure. Other Linux entry points without --pixel may use the
Hermes fallback on unsupported hosts. See docs/PIXEL.md. Hermes-enabled installs
keep the bootstrap model at a 64K context floor, then promote the full local
model target to 128K after the background swap.
Check progress:
ods status
tail -f ~/ods/logs/model-upgrade.log
On Windows:
cd $env:USERPROFILE\ods
.\ods.ps1 status
Get-Content .\logs\model-upgrade.log -Wait
Loopback-only installs open the Chat UI directly without an account. A
network-bound or ODS proxy install keeps authentication enabled and prompts the
first user to create the admin account.
Validate The Install
ods status
ods chat "Say exactly: ODS is ready."
ods doctor
On Windows:
cd $env:USERPROFILE\ods
.\ods.ps1 status
.\ods.ps1 logs llm
.\ods.ps1 report
For a lower-level source-tree check on Linux/macOS, or inside Ubuntu for Windows/WSL:
cd~/ods
./ods-preflight.sh
./scripts/ods-test.sh
Test The Local API
Use the port written to .env. Linux Docker installs commonly expose
llama-server on OLLAMA_PORT=11434; macOS native Metal installs commonly use
8080.