Skip to content
CLIOSearchCtrl K
Download

Managed infrastructure

From the Infrastructure page, CLIO can install, start, stop, check and remove the services your work needs:

  • model servers: vLLM, llama.cpp and Ollama
  • a model router: LiteLLM Proxy
  • provenance capture: Flowcept and HPE CMF
  • web search: CLIO Web Search and SearXNG

It installs them on the computer running CLIO or on a Linux server you reach over SSH, such as a cluster node. From the same page CLIO can also install CLIO itself on a Linux server over SSH, and the CLIO Relay tool on a cluster.

Each service offers one or more options. The Infrastructure page shows the ones that fit the computer you selected.

  • Installs directly. CLIO downloads the service’s official release, a ready-to-run program pinned to an exact version and checked against a known checksum, into its own folder. Root access and a container runtime are not needed.
  • Container. CLIO runs the service’s official image, pinned to an exact version, with Docker or Podman, or with Apptainer on Linux clusters. For Flowcept and HPE CMF, the application installs directly and only its databases run in containers.

CLIO does not build software from source, so you never need a compiler. Before a direct install finishes, CLIO starts the downloaded program once. If the program cannot run on that computer, for example because the operating system is too old, CLIO stops the install and names what is missing. Choose a container option instead.

Each table lists the options CLIO offers for one kind of computer, with the version it installs. “Installs directly” needs nothing beyond what the next section lists. The last column shows what a container runtime adds. Linux rows apply to the computer running CLIO and to servers you reach over SSH. On Windows and macOS, services install on the computer running CLIO.

Service Installs directly Also available with Docker, Podman or Apptainer
vLLM Native CUDA (0.28.0); Native CUDA + attention (0.27.0) NVIDIA CUDA (0.28.0); CPU (0.28.0)
Model router LiteLLM Proxy (native) (1.102.2) Nothing more
llama.cpp NVIDIA CUDA (native) (b11206); CPU (native) (b11206); Vulkan (AMD, Intel or NVIDIA) (native) (b11206) NVIDIA CUDA (b11206); Vulkan (b11206); CPU (b11206)
Ollama NVIDIA CUDA (native) (0.34.4); CPU (native) (0.34.4); Vulkan (AMD, Intel or NVIDIA) (native) (0.34.4) CPU (0.34.4); NVIDIA CUDA (0.34.4)
Flowcept Native application + native/site databases (e638b4e2) Native application + container databases (1.0.3)
HPE CMF Native application + native/site databases (53d9c3e5) Native application + container databases (53d9c3e5)
CLIO Web Search Native gateway + SearXNG (0.3.2) Container (0.3.2)
SearXNG Native (uv) (2026.8.11+e8e710e42) Nothing more
Service Installs directly Also available with Docker, Podman or Apptainer
vLLM Not available AMD ROCm (0.28.0); CPU (0.28.0)
Model router LiteLLM Proxy (native) (1.102.2) Nothing more
llama.cpp CPU (native) (b11206); Vulkan (AMD, Intel or NVIDIA) (native) (b11206) Vulkan (b11206); CPU (b11206)
Ollama AMD ROCm (native) (0.34.4); CPU (native) (0.34.4); Vulkan (AMD, Intel or NVIDIA) (native) (0.34.4) CPU (0.34.4); AMD ROCm (0.34.4)
Flowcept Native application + native/site databases (e638b4e2) Native application + container databases (1.0.3)
HPE CMF Native application + native/site databases (53d9c3e5) Native application + container databases (53d9c3e5)
CLIO Web Search Native gateway + SearXNG (0.3.2) Container (0.3.2)
SearXNG Native (uv) (2026.8.11+e8e710e42) Nothing more
Service Installs directly Also available with Docker, Podman or Apptainer
vLLM Native Intel XPU (host); Native Intel XPU + attention (host) CPU (0.28.0)
Model router LiteLLM Proxy (native) (1.102.2) Nothing more
llama.cpp Vulkan (AMD, Intel or NVIDIA) (native) (b11206); CPU (native) (b11206) CPU (b11206)
Ollama Vulkan (AMD, Intel or NVIDIA) (native) (0.34.4); CPU (native) (0.34.4) CPU (0.34.4)
Flowcept Native application + native/site databases (e638b4e2) Native application + container databases (1.0.3)
HPE CMF Native application + native/site databases (53d9c3e5) Native application + container databases (53d9c3e5)
CLIO Web Search Native gateway + SearXNG (0.3.2) Container (0.3.2)
SearXNG Native (uv) (2026.8.11+e8e710e42) Nothing more
Service Installs directly Also available with Docker, Podman or Apptainer
vLLM Not available CPU (0.28.0)
Model router LiteLLM Proxy (native) (1.102.2) Nothing more
llama.cpp CPU (native) (b11206) CPU (b11206)
Ollama CPU (native) (0.34.4) CPU (0.34.4)
Flowcept Native application + native/site databases (e638b4e2) Native application + container databases (1.0.3)
HPE CMF Native application + native/site databases (53d9c3e5) Native application + container databases (53d9c3e5)
CLIO Web Search Native gateway + SearXNG (0.3.2) Container (0.3.2)
SearXNG Native (uv) (2026.8.11+e8e710e42) Nothing more
Service Installs directly Also available with Docker, Podman or Apptainer
vLLM Native CUDA (0.28.0); Native CUDA + attention (0.27.0) NVIDIA CUDA (0.28.0)
Model router LiteLLM Proxy (native) (1.102.2) Nothing more
llama.cpp CPU (native) (b11206); Vulkan (AMD, Intel or NVIDIA) (native) (b11206) NVIDIA CUDA (b11206); Vulkan (b11206); CPU (b11206)
Ollama NVIDIA CUDA (native) (0.34.4); CPU (native) (0.34.4); Vulkan (AMD, Intel or NVIDIA) (native) (0.34.4) CPU (0.34.4); NVIDIA CUDA (0.34.4)
Flowcept Not available Native application + container databases (1.0.3)
HPE CMF Not available Native application + container databases (53d9c3e5)
CLIO Web Search Native gateway + SearXNG (0.3.2) Container (0.3.2)
SearXNG Native (uv) (2026.8.11+e8e710e42) Nothing more
Service Installs directly Also available with Docker, Podman or Apptainer
vLLM Not available Nothing more
Model router LiteLLM Proxy (native) (1.102.2) Nothing more
llama.cpp CPU (native) (b11206) CPU (b11206)
Ollama CPU (native) (0.34.4) CPU (0.34.4)
Flowcept Not available Native application + container databases (1.0.3)
HPE CMF Not available Native application + container databases (53d9c3e5)
CLIO Web Search Native gateway + SearXNG (0.3.2) Container (0.3.2)
SearXNG Native (uv) (2026.8.11+e8e710e42) Nothing more
Service Installs directly Also available with Docker, Podman or Apptainer
vLLM Not available CPU (0.28.0)
Model router Not available Nothing more
llama.cpp NVIDIA CUDA (native) (b11206); CPU (native) (b11206); Vulkan (AMD, Intel or NVIDIA) (native) (b11206) CPU (b11206)
Ollama NVIDIA CUDA (native) (0.34.4); CPU (native) (0.34.4); Vulkan (AMD, Intel or NVIDIA) (native) (0.34.4) CPU (0.34.4)
Flowcept Not available Nothing more
HPE CMF Not available Nothing more
CLIO Web Search Native gateway + SearXNG (0.3.2) Container (0.3.2)
SearXNG Native (uv) (2026.8.11+e8e710e42) Nothing more
Service Installs directly Also available with Docker, Podman or Apptainer
vLLM Not available CPU (0.28.0)
Model router Not available Nothing more
llama.cpp Vulkan (AMD, Intel or NVIDIA) (native) (b11206); CPU (native) (b11206); AMD ROCm (native) (b11206) CPU (b11206)
Ollama Vulkan (AMD, Intel or NVIDIA) (native) (0.34.4); CPU (native) (0.34.4); AMD ROCm (native) (0.34.4) CPU (0.34.4)
Flowcept Not available Nothing more
HPE CMF Not available Nothing more
CLIO Web Search Native gateway + SearXNG (0.3.2) Container (0.3.2)
SearXNG Native (uv) (2026.8.11+e8e710e42) Nothing more
Service Installs directly Also available with Docker, Podman or Apptainer
vLLM Not available CPU (0.28.0)
Model router Not available Nothing more
llama.cpp CPU (native) (b11206) CPU (b11206)
Ollama CPU (native) (0.34.4) CPU (0.34.4)
Flowcept Not available Nothing more
HPE CMF Not available Nothing more
CLIO Web Search Native gateway + SearXNG (0.3.2) Container (0.3.2)
SearXNG Native (uv) (2026.8.11+e8e710e42) Nothing more
Service Installs directly Also available with Docker, Podman or Apptainer
vLLM Not available Nothing more
Model router Not available Nothing more
llama.cpp CPU (native) (b11206) CPU (b11206)
Ollama CPU (native) (0.34.4) CPU (0.34.4)
Flowcept Not available Nothing more
HPE CMF Not available Nothing more
CLIO Web Search Not available Container (0.3.2)
SearXNG Not available Nothing more
Service Installs directly Also available with Docker, Podman or Apptainer
vLLM Not available Nothing more
Model router LiteLLM Proxy (native) (1.102.2) Nothing more
llama.cpp Apple Metal (native) (b11206) CPU (b11206)
Ollama Apple Metal (native) (0.34.4) CPU (0.34.4)
Flowcept Not available Nothing more
HPE CMF Not available Nothing more
CLIO Web Search Native gateway + SearXNG (0.3.2) Container (0.3.2)
SearXNG Native (uv) (2026.8.11+e8e710e42) Nothing more
Service Installs directly Also available with Docker, Podman or Apptainer
vLLM Not available CPU (0.28.0)
Model router LiteLLM Proxy (native) (1.102.2) Nothing more
llama.cpp CPU (native) (b11206) CPU (b11206)
Ollama CPU (native) (0.34.4) CPU (0.34.4)
Flowcept Not available Nothing more
HPE CMF Not available Nothing more
CLIO Web Search Native gateway + SearXNG (0.3.2) Container (0.3.2)
SearXNG Native (uv) (2026.8.11+e8e710e42) Nothing more

Direct installs on Linux use the official Linux releases. These are built for current distributions, so they need a C library (glibc) of version 2.34 or newer and a few common system libraries.

Service part Needs
Valkey, Flowcept’s message queue glibc 2.34 or newer, OpenSSL 3 (libssl.so.3), libsystemd.so.0
MongoDB, Flowcept’s database Ubuntu 20.04, 22.04 or 24.04, Debian 12, RHEL, Rocky or AlmaLinux 8, or SLES or openSUSE Leap 15. SLES and Leap also need the libopenssl1_1 package.
PostgreSQL, HPE CMF’s database glibc 2.34 or newer, OpenSSL 3, libxml2, Kerberos (libgssapi_krb5), lz4 and zstd
llama.cpp glibc 2.34 or newer, the C++ library from GCC 12 or newer, and the OpenMP runtime (libgomp.so.1). The NVIDIA CUDA option also needs the NVIDIA driver; CLIO downloads the CUDA 12.8 runtime it uses.

llama.cpp publishes no Linux build for AMD ROCm. On AMD GPUs, use the Vulkan option or a container.

These distributions were checked by starting each program in the distribution’s official image, after installing the libraries above:

Distribution glibc Valkey PostgreSQL llama.cpp
Ubuntu 24.04 2.39 Runs Runs Runs
Ubuntu 22.04 2.35 Runs Runs Runs
openSUSE Leap 15.6, SLES 15 SP6 2.38 Runs Runs Runs
Rocky Linux 9, RHEL 9 2.34 Runs Runs Does not run: its C++ library is too old
openSUSE Leap 15.5, SLES 15 SP5 2.31 Does not run Does not run Does not run
Rocky Linux 8, RHEL 8 2.28 Does not run Does not run Does not run

Direct installs run on the Windows computer running CLIO. To use a Windows server, run CLIO on it.

  • llama.cpp needs the Microsoft Visual C++ 2015–2022 Redistributable (x64), which most Windows computers already have. If it is missing, CLIO stops the install and says so.
  • The llama.cpp NVIDIA CUDA option needs the NVIDIA driver. CLIO downloads the CUDA 12.4 runtime and cuBLAS that llama.cpp publishes with it.
  • Flowcept and HPE CMF do not run on Windows. Install them on a Linux server.
  • The CLIO desktop app and Ollama need macOS 14 or newer. llama.cpp needs macOS 13.3 or newer.
  • On Apple silicon, llama.cpp and Ollama install directly and use the GPU through Metal. On Intel Macs they run on the CPU.
  • vLLM does not install directly on macOS. On an Intel Mac with Docker or Podman, its CPU container is available.
  • Flowcept and HPE CMF do not run on macOS. Install them on a Linux server.