Installation

Inclavate ships as a self-contained bundle — the dppan CLI, the orchestrator, and a node binary. The quickest path is the one-line installer; a manual download is available too.

Prerequisites

Ollama provides the model weights (GGUF). Install it and pull a model:

# macOS / Linux
curl -fsSL https://ollama.com/install.sh | sh
ollama pull llama3.2:1b

On Windows, download the installer from ollama.com, then run ollama pull llama3.2:1b.

Platform requirements

Platform	Minimum	GPU
macOS	13 Ventura · Apple Silicon (M1+)	Metal
Linux	glibc 2.31+ (Ubuntu 20.04+, Debian 11+)	NVIDIA driver 520+ / CUDA 12 (optional)
Windows	10 21H2 / 11 (64-bit) · VC++ 2022 redist	NVIDIA driver 576.02+ / CUDA 12 (optional)

GPU is optional — CPU fallback works everywhere.

Supported GPUs

The prebuilt CUDA binaries include kernels for a fixed set of NVIDIA architectures:

GPU	Linux	Windows
GTX 16xx · RTX 20 / 30 / 40 series	✅ native	✅ native
H100 / H200 (Hopper)	✅ native	⚡ JIT¹
RTX 50xx / Blackwell	⚡ JIT¹	⚡ JIT¹
GTX 10xx (Pascal) · A100 / A30	❌ not supported²	❌ not supported²

¹ Runs, but the driver compiles the kernels on the first launch (a few extra seconds), then caches them for later runs. ² Build from source with the matching architecture, e.g. -DCMAKE_CUDA_ARCHITECTURES="61" (Pascal) or "80" (A100).

On macOS, Metal acceleration is automatic on Apple Silicon (M1+); Intel Macs run CPU-only.

Quick install

macOS / Linux

curl -fsSL https://dl.inclavate.io/install.sh | sh

Windows · PowerShell

irm https://dl.inclavate.io/install.ps1 | iex

This installs the latest release under ~/.inclavate and adds ~/.inclavate/bin to your PATH — no sudo required. Open a new shell (or source your shell rc), then verify:

dppan --version

Pin a specific version:

curl -fsSL https://dl.inclavate.io/install.sh | sh -s -- --version v0.7.4

What the installer does

1. Detects your OS and CPU architecture and picks the matching release.
2. Downloads the archive and verifies its SHA-256 checksum before touching anything.
3. Extracts the whole bundle to ~/.inclavate/versions/<version>/ (macOS/Linux) or %LOCALAPPDATA%\Inclavate\versions\<version>\ (Windows) — the binaries, their shared libraries, frontend/, and config/ all stay together (they resolve relative to each other).
4. Links the three binaries — dppan, dppan-node, dppan-orchestrator — into …/bin (symlinks on macOS/Linux, .cmd shims on Windows).
5. Adds that bin directory to your user PATH (your shell rc on macOS/Linux, the User environment on Windows). No administrator / sudo rights are used — nothing is written outside your home directory.

The PATH change applies to new shells — open a fresh terminal (or source your rc) afterwards.

Options

macOS / Linux (install.sh)	Windows (install.ps1)	Effect
--version vX.Y.Z	-Version vX.Y.Z	Install a specific release instead of the latest
--cuda / --cpu	—	Force the Linux acceleration build (skip the prompt)
--prefix DIR	-Prefix DIR	Install to a custom root (defaults above)
--no-modify-path	-NoModifyPath	Install without editing PATH; print the line to add yourself
--uninstall	-Uninstall	Remove the install and its PATH entry
-y / --yes	—	Non-interactive: take defaults, never prompt

Pass flags through the one-liner with sh -s --:

curl -fsSL https://dl.inclavate.io/install.sh | sh -s -- --cpu --prefix ~/tools/inclavate

Platform & acceleration defaults

CPU execution works everywhere; GPU acceleration is selected automatically where available.

Platform	Default build	GPU acceleration	Asks?
macOS · Apple Silicon	universal	Metal — always on	no
Linux · x86_64	CPU	opt-in CUDA build	yes, if nvidia-smi is found
Linux · arm64	CPU	— (no CUDA build)	no
Windows · x86_64	CUDA build	CUDA if a supported GPU + driver is present, else CPU fallback	no

On Linux x86_64, if an NVIDIA GPU is detected the installer asks whether to install the CUDA build (GPU-accelerated) or the CPU build. --cuda / --cpu skip the question, and non-interactive runs (CI, or piped with no terminal) default to CPU. With no accelerator it installs the CPU build and notes that inference/compute may be slower. See Supported GPUs for CUDA architecture coverage.

Manual download

Grab the build for your platform from inclavate.io/downloads, then run the bundled installer:

macOS / Linux

tar -xzf dppan-<version>-macos-arm64.tar.gz
cd dppan-<version>-macos-arm64
./install.sh              # installs to ~/.inclavate + PATH, no network

Windows · PowerShell

Expand-Archive dppan-<version>-windows-x86_64.zip .
cd dppan-<version>-windows-x86_64
.\install.ps1

Or skip installing entirely and run it straight from the extracted folder — the binaries locate their libraries relative to themselves:

./dppan --version

Updating

Once installed, update in place:

dppan update

It checks for a newer release and, if there is one, re-runs the installer to fetch it and repoint the dppan, dppan-node, and dppan-orchestrator symlinks. Pin a version with dppan update --version v0.7.4, or reinstall the current one with --force.

Uninstall

dppan uninstall          # add -y to skip the confirmation prompt

This strips the PATH entry the installer added and removes the install directory. If dppan isn't on your PATH, the installer's own flag does the same thing:

macOS / Linux

curl -fsSL https://dl.inclavate.io/install.sh | sh -s -- --uninstall

Windows · PowerShell

& ([scriptblock]::Create((irm https://dl.inclavate.io/install.ps1))) -Uninstall

You're ready — continue to Usage →.