Scout on-premise setup

Run the Scout AI agent in your own infrastructure, with your own LLM endpoint

​

Prerequisites

To run Scout on-premise, you need:

• A place to run it: a Kubernetes cluster, a container runtime, or a host for the binary
• OpenAI-compatible API (e.g. LiteLLM) serving an Anthropic Claude model
• Coalesce Quality client credentials (QUALITY_CLIENT_ID and QUALITY_CLIENT_SECRET), see Configuration

​

Deployment

Scout runs as a single process that makes only outbound connections. You can run it inside your network without exposing any port.

Kubernetes

Use Kubernetes for a long-running deployment. The manifests, a Kustomize overlay, setup instructions and a worked example are in getsynq/synq-scout-k8s.

Binary

Use the binary to try Scout, or to run it on a host that you manage.

On macOS, run brew install getsynq/tap/synq-scout, and brew upgrade to upgrade. For a long-running deployment, use the container image or the Kubernetes manifests instead.

Otherwise, download an archive from the releases page. The file name contains the version. The first line below gets the latest version from GitHub:

VERSION=$(curl -fsSLI -o /dev/null -w '%{url_effective}' \
  https://github.com/getsynq/synq-scout/releases/latest | sed 's#.*/v##')
OS=$(uname -s | tr '[:upper:]' '[:lower:]')          # darwin or linux
ARCH=$(uname -m | sed 's/x86_64/amd64/; s/aarch64/arm64/')

curl -fL "https://github.com/getsynq/synq-scout/releases/download/v${VERSION}/synq-scout_${VERSION}_${OS}_${ARCH}.tar.gz" \
  | tar -xz
sudo mv synq-scout /usr/local/bin/

Builds are published for macOS, Linux and Windows, each on both amd64 and arm64, and every release ships a checksums.txt you can verify with sha256sum -c checksums.txt --ignore-missing. Confirm which build you have with synq-scout --version.

On Windows, download synq-scout_<version>_windows_amd64.zip (or _arm64) from the same page and extract synq-scout.exe to a directory on your PATH:

Expand-Archive synq-scout_<version>_windows_amd64.zip -DestinationPath "$env:LOCALAPPDATA\synq-scout"
$env:Path += ";$env:LOCALAPPDATA\synq-scout"   # this session; add it to your user PATH to keep it
synq-scout --version

Scout connects to every supported warehouse except DuckDB.

Each archive contains an agent.example.yaml to start from. Check the setup before you run the agent:

synq-scout health

Then start the agent:

synq-scout agent

Docker

docker pull europe-docker.pkg.dev/synq-cicd-public/synq-public/synq-scout:latest

This is the image that the Kubernetes manifests deploy. Use it to run Scout in your own container runtime.

To use the Scout tools from an AI client, you do not need an on-premise deployment. Use the hosted MCP server, see MCP.

​

Configuration

Scout signs in to Coalesce Quality with a client ID and a client secret. To get them:

1. Go to Settings → Scout AI → Infrastructure.
2. Set the deployment method to Self-hosted (on-premises).
3. Copy the client ID and client secret shown under Self-hosted infrastructure.

Give them to Scout as the QUALITY_CLIENT_ID and QUALITY_CLIENT_SECRET environment variables, or in the synq block of agent.yaml. If your workspace is not in the EU region, also set QUALITY_REGION to us or au.

The Scout CLI reference lists the other ways to sign in, such as synq-scout auth login.

​

LLM Model Configuration

Scout sends its model calls to an OpenAI-compatible endpoint, such as a LiteLLM proxy. It uses two models: a thinking model for reasoning and triage, and a summary model for summaries. If you do not set them, both are Claude Sonnet.

Set the endpoint and the models with these environment variables, or in the llm.openai block of agent.yaml:

Variable	Description
OPENAI_BASE_URL	URL of the OpenAI-compatible endpoint. LITELLM_BASE_URL also works
OPENAI_API_KEY	API key for the endpoint. LITELLM_API_KEY and LITELLM_MASTER_KEY also work
QUALITY_THINKING_MODEL	Model used for reasoning and triage
QUALITY_SUMMARY_MODEL	Model used for summaries
OPENAI_MODEL	Model for both roles, if the two variables above are not set

The model names should match what your OpenAI-compatible API expects (e.g. the model name configured in LiteLLM).

To see the supported models and their stable and latest versions, go to Settings → Scout AI → Models in Coalesce Quality.

​

Support

For help with a deployment, read the synq-scout-k8s repository, report a problem in its issues, or contact your Coalesce Quality team.