Sherpa AI Server#

Download the distribution#

Choose one published Sherpa AI Server version. The client-files archive determines the Docker image version for installation or updating. Components should not be downloaded from separate /latest links, as a partial release can mix different versions.

This example uses version 2.5.1. Replace it with the published version you need and run the commands in the installation directory:

VERSION=2.5.1
CLIENT_ARCHIVE="docker_aiserver_client_files_${VERSION}.tgz"
BASE_URL="https://downloads.sherparpa.ru/SherpaAIServer/aiserver-client-files"

curl -fL -O "${BASE_URL}/${CLIENT_ARCHIVE}"
curl -fL -O "${BASE_URL}/${CLIENT_ARCHIVE}.SHA256SUM"

EXPECTED_SHA256="$(awk '{print $1}' "${CLIENT_ARCHIVE}.SHA256SUM")"
ACTUAL_SHA256="$(sha256sum "${CLIENT_ARCHIVE}" | awk '{print $1}')"
test "$ACTUAL_SHA256" = "$EXPECTED_SHA256" && echo 'SHA-256 OK'

If the check does not print SHA-256 OK, do not extract the archive. Check the version and download the files again. The file command identifies the archive type but does not verify integrity.

Install with access to the download server#

After verifying the archive, extract it and run the installer:

tar -xzf "$CLIENT_ARCHIVE"
chmod +x sh_scripts/*.sh
sudo ./sh_scripts/install.sh

The installer selects images according to the client-files version, offers required models, and checks archive SHA-256 values and Docker manifest tags before starting services. The image set depends on CPU or GPU, Whisper and reranker options. Models are downloaded separately and are not tied to a release version.

Follow the installation guide for detailed steps. For an existing installation, use the update guide.

Install without access to the download server#

On a machine that can reach the distribution, obtain client-files, the required Docker image archives of exactly the same version, and each <archive name>.SHA256SUM file. Move them into one directory on the target server and run install.sh. It uses and verifies local files before installation. Obtain any missing archives or checksums before continuing.

The core components are aiserver, aiserver-pg, aiserver-embed, aiserver-nginx, aiserver-websocket, aiserver-code-interpreter, aiserver-mcp-orchestrator and mypy. GPU also requires aiserver-llm-server. CPU requires aiserver-llm-server-cpu. aiserver-whisper and aiserver-bge-reranker depend on the selected features. Image archives use the name <component>_<version>.tar.gz.

Transfer files with SCP/SFTP or removable media. After copying, check free space for archives, temporary files, extraction and running containers:

df -h .
ls -lh docker_aiserver_client_files_*.tgz *.tar.gz *.SHA256SUM

Do not manually stop a running installation while preparing files: the update script checks availability, version, checksums, Docker manifest and free space before showing its action plan.