scalattice/scalattice-agent

Open-source GPU operator agent for the Scalattice inference network (Rust).

Rust

9

572 commits

updated Sep 15, 2026

See the code
ai
compute
compute-cluster
compute-clusters
gpu
gpu-cluster
hypervisor
inference

README

Scalattice

Scalattice Agent

Open-source GPU agent for the Scalattice inference network, a product of Robottik Ltd.

Install it on a machine with a GPU, connect with a provider token, and Scalattice Cloud can route inference jobs to you.

Install

Linux

curl -fsSL https://scalattice.cloud/install/agent | sh -s -- --token slt_provider_YOUR_TOKEN
source ~/.config/scalattice/agent.env
scalattice-agent status

Full agent logs go to ~/.local/share/scalattice/agent.log (always verbose; rotated at 8 MiB with one backup). scalattice-agent logs tails that file.

Windows

Download and run ScalatticeAgentSetup-x86_64.exe from GitHub Releases (or use Download Windows installer on the Providers dashboard).

  1. Run the setup wizard (approve SmartScreen if prompted)
  2. Paste your slt_provider_… token when asked
  3. Finish — PATH is configured and the background agent starts

Setup guide: https://scalattice.cloud/install/agent-setup

The agent registers a boot scheduled task (runs as SYSTEM before sign-in) plus logon Startup/Task rails. Use the notification area icon for status, token changes, and live logs. Full detail is written to %LOCALAPPDATA%\Scalattice\logs\agent.log (rotated at 8 MiB, one backup). The tray Live log can show a Simplified view; the file itself is always verbose.

macOS (Apple Silicon)

Release builds are arm64 + Metal only. Intel Macs are not supported.

Download and open ScalatticeAgentSetup-aarch64.dmg (or use Download macOS installer on the Providers dashboard).

  1. Open the disk image and run Scalattice Agent (approve Gatekeeper if prompted)
  2. Paste your slt_provider_… token in the control panel
  3. Finish — the agent stays in the menu bar

Setup guide: https://scalattice.cloud/install/agent-setup?os=macos

The binary is compiled and signed on GitHub’s macos-14 runners. See docs/macos-signing.md.

Background mode uses a LaunchAgent. Stay signed in (or enable auto-login) so launchd keeps the agent running. The menu bar app is the same control panel as Windows (status, token, logs).

Uninstall

On Windows, use Settings → Apps. The wizard asks whether to delete downloaded models; logs and settings are always removed.

On Linux and macOS:

scalattice-agent uninstall

That prompts to keep or delete model weights, then asks you to type yes. Use --yes to skip the confirmation (models are kept) and --purge to delete models too.

Releases are self-contained binaries (plus bundled runtime libraries where allowed). You can also build from source with Rust/Cargo.

Hardware support

Release builds include CUDA (NVIDIA) and Vulkan (AMD/Intel/ARM) on Linux and Windows, and Metal on Apple Silicon Macs.

HardwareWhat the host needs
NVIDIA GTX/RTXNVIDIA driver installed (nvidia-smi works)
AMD / Intel GPUVendor Vulkan ICD (AMD Adrenalin / Intel graphics driver). ROCm is not required. Discrete AMD/Intel are detected on Linux (lspci) and Windows (Win32 video controllers) and run through Vulkan when enabled in the Providers dashboard.
Apple SiliconmacOS 14+, M1 or later. Unified memory is advertised as GPU VRAM (minus OS headroom). Intel Macs are not supported.
CPU onlyNothing extra — the agent connects; inference uses the CPU backend

When both NVIDIA and AMD/Intel are enabled, the agent prefers CUDA. AMD/Intel-only pools use the same load cascade as NVIDIA (gpu-full → offload → CPU). Check agent logs for Vulkan pool: pinning ggml backend device(s).

Do not redistribute vulkan-1.dll — providers get it from their GPU driver. Build machines need the Vulkan SDK (for glslc only).

NVIDIA’s libcuda.so / nvcuda.dll cannot be redistributed. Machines without a GPU driver still run the agent; GPU inference activates when the driver is present.

Quick start

  1. Sign in at scalattice.cloudProviders
  2. Register as a provider and create an agent token (slt_provider_…)
  3. On your GPU machine:
curl -fsSL https://scalattice.cloud/install/agent | sh -s -- --token slt_provider_…
source ~/.config/scalattice/agent.env
scalattice-agent status

The install script saves your token and starts the background agent. To change the token later:

scalattice-agent set-token --token slt_provider_…

Environment

VariableDescription
SCALATTICE_AGENT_TOKENProvider agent token (slt_provider_…)

The agent connects to Scalattice Cloud using the built-in endpoint. Region, models, and schedule are managed in the Providers dashboard — not via local flags.

Protocol

See docs/AGENT_PROTOCOL.md.

Build from source

# x86_64 Linux (NVIDIA + AMD/Intel Vulkan)
cargo build --release --features gpu

# aarch64 Linux (NVIDIA + ARM Vulkan) — build natively on ARM
cargo build --release --no-default-features --features arm-gpu
# x86_64 Windows (NVIDIA CUDA + AMD/Intel Vulkan) — run in PowerShell
# Requires Vulkan SDK on the build machine (glslc). Providers do not need the SDK.
.\scripts\build-release.ps1

CI release binaries: CUDA + Vulkan on x86_64/aarch64 Linux and x86_64 Windows.

Community

DocumentPurpose
CONTRIBUTING.mdHow to report issues and when we accept changes
SECURITY.mdResponsible disclosure for security issues
CODE_OF_CONDUCT.mdCommunity standards

License

MIT. See LICENSE.

Contributors

romulushill

495 commits

scalattice/scalattice-agent

Open-source GPU operator agent for the Scalattice inference network (Rust).

Rust

9

572 commits

updated Sep 15, 2026

See the code
ai
compute
compute-cluster
compute-clusters
gpu
gpu-cluster
hypervisor
inference

README

Scalattice

Scalattice Agent

Open-source GPU agent for the Scalattice inference network, a product of Robottik Ltd.

Install it on a machine with a GPU, connect with a provider token, and Scalattice Cloud can route inference jobs to you.

Install

Linux

curl -fsSL https://scalattice.cloud/install/agent | sh -s -- --token slt_provider_YOUR_TOKEN
source ~/.config/scalattice/agent.env
scalattice-agent status

Full agent logs go to ~/.local/share/scalattice/agent.log (always verbose; rotated at 8 MiB with one backup). scalattice-agent logs tails that file.

Windows

Download and run ScalatticeAgentSetup-x86_64.exe from GitHub Releases (or use Download Windows installer on the Providers dashboard).

  1. Run the setup wizard (approve SmartScreen if prompted)
  2. Paste your slt_provider_… token when asked
  3. Finish — PATH is configured and the background agent starts

Setup guide: https://scalattice.cloud/install/agent-setup

The agent registers a boot scheduled task (runs as SYSTEM before sign-in) plus logon Startup/Task rails. Use the notification area icon for status, token changes, and live logs. Full detail is written to %LOCALAPPDATA%\Scalattice\logs\agent.log (rotated at 8 MiB, one backup). The tray Live log can show a Simplified view; the file itself is always verbose.

macOS (Apple Silicon)

Release builds are arm64 + Metal only. Intel Macs are not supported.

Download and open ScalatticeAgentSetup-aarch64.dmg (or use Download macOS installer on the Providers dashboard).

  1. Open the disk image and run Scalattice Agent (approve Gatekeeper if prompted)
  2. Paste your slt_provider_… token in the control panel
  3. Finish — the agent stays in the menu bar

Setup guide: https://scalattice.cloud/install/agent-setup?os=macos

The binary is compiled and signed on GitHub’s macos-14 runners. See docs/macos-signing.md.

Background mode uses a LaunchAgent. Stay signed in (or enable auto-login) so launchd keeps the agent running. The menu bar app is the same control panel as Windows (status, token, logs).

Uninstall

On Windows, use Settings → Apps. The wizard asks whether to delete downloaded models; logs and settings are always removed.

On Linux and macOS:

scalattice-agent uninstall

That prompts to keep or delete model weights, then asks you to type yes. Use --yes to skip the confirmation (models are kept) and --purge to delete models too.

Releases are self-contained binaries (plus bundled runtime libraries where allowed). You can also build from source with Rust/Cargo.

Hardware support

Release builds include CUDA (NVIDIA) and Vulkan (AMD/Intel/ARM) on Linux and Windows, and Metal on Apple Silicon Macs.

HardwareWhat the host needs
NVIDIA GTX/RTXNVIDIA driver installed (nvidia-smi works)
AMD / Intel GPUVendor Vulkan ICD (AMD Adrenalin / Intel graphics driver). ROCm is not required. Discrete AMD/Intel are detected on Linux (lspci) and Windows (Win32 video controllers) and run through Vulkan when enabled in the Providers dashboard.
Apple SiliconmacOS 14+, M1 or later. Unified memory is advertised as GPU VRAM (minus OS headroom). Intel Macs are not supported.
CPU onlyNothing extra — the agent connects; inference uses the CPU backend

When both NVIDIA and AMD/Intel are enabled, the agent prefers CUDA. AMD/Intel-only pools use the same load cascade as NVIDIA (gpu-full → offload → CPU). Check agent logs for Vulkan pool: pinning ggml backend device(s).

Do not redistribute vulkan-1.dll — providers get it from their GPU driver. Build machines need the Vulkan SDK (for glslc only).

NVIDIA’s libcuda.so / nvcuda.dll cannot be redistributed. Machines without a GPU driver still run the agent; GPU inference activates when the driver is present.

Quick start

  1. Sign in at scalattice.cloudProviders
  2. Register as a provider and create an agent token (slt_provider_…)
  3. On your GPU machine:
curl -fsSL https://scalattice.cloud/install/agent | sh -s -- --token slt_provider_…
source ~/.config/scalattice/agent.env
scalattice-agent status

The install script saves your token and starts the background agent. To change the token later:

scalattice-agent set-token --token slt_provider_…

Environment

VariableDescription
SCALATTICE_AGENT_TOKENProvider agent token (slt_provider_…)

The agent connects to Scalattice Cloud using the built-in endpoint. Region, models, and schedule are managed in the Providers dashboard — not via local flags.

Protocol

See docs/AGENT_PROTOCOL.md.

Build from source

# x86_64 Linux (NVIDIA + AMD/Intel Vulkan)
cargo build --release --features gpu

# aarch64 Linux (NVIDIA + ARM Vulkan) — build natively on ARM
cargo build --release --no-default-features --features arm-gpu
# x86_64 Windows (NVIDIA CUDA + AMD/Intel Vulkan) — run in PowerShell
# Requires Vulkan SDK on the build machine (glslc). Providers do not need the SDK.
.\scripts\build-release.ps1

CI release binaries: CUDA + Vulkan on x86_64/aarch64 Linux and x86_64 Windows.

Community

DocumentPurpose
CONTRIBUTING.mdHow to report issues and when we accept changes
SECURITY.mdResponsible disclosure for security issues
CODE_OF_CONDUCT.mdCommunity standards

License

MIT. See LICENSE.

Contributors

romulushill

495 commits

Languages

Rust

76.4%

PowerShell

8.8%

Python

7.9%

Inno Setup

3.2%

Shell

2.9%