Get DeltaForge on your machine

Four packages, one native binary each. No Spark, no JVM. Pick the package for your workflow below, or deploy the full platform to Azure, AWS, or Google Cloud with one template.

Signed binaries for every platform: deltaforge-org/delta-forge on GitHub

Install the desktop application

The interactive query editor, catalog browser, and pipeline viewer. A single signed binary for macOS, Windows, and Linux: no Spark, no JVM.

curl -fsSL https://deltaforge.org/install.sh | sh

Or via Homebrew:

# Add the tap (one-time)
brew tap deltaforge-org/tap
brew install --cask deltaforge-platform
PowerShell
iwr -useb https://deltaforge.org/install.ps1 | iex

Or via winget (installs the signed MSI and adds the application to your PATH):

winget install DeltaForge.Platform

Detects your architecture (x64 or ARM64), downloads the matching AppImage, and adds a launcher entry:

curl -fsSL https://deltaforge.org/install.sh | sh

Prefer the terminal? Switch to deltaforge-cli above for scripting and CI, or deltaforge-compute to run a production node.

Verify the download

Every release artifact includes a detached GPG signature and a SHA256SUMS manifest.

# Import the signing key (one-time)
curl -fsSL https://deltaforge.org/pubkey.asc | gpg --import

# Verify a downloaded artifact
gpg --verify deltaforge-cli-1.0.7-linux-x64.tar.gz.sig \
             deltaforge-cli-1.0.7-linux-x64.tar.gz

Fingerprint: B461 8130 D4B0 3CF5 454E  28F0 A859 0BF7 C3DC E5F3  ·  RSA 4096, UID releases@deltaforge.org

FAQ

No. Every DeltaForge package is a single native binary. There is no Spark cluster to stand up and no JVM to configure: install the package for your platform and start querying Delta Lake and Iceberg tables immediately.
Use the deltaforge desktop application for interactive development on macOS, Windows, or Linux. For terminal and CI work, switch to deltaforge-cli. For cloud production deployments, use deltaforge-compute.
Switch to Cloud deployment above: one native template per cloud (Bicep, CloudFormation, Terraform) provisions the whole platform, including the PostgreSQL catalog, object storage, and HTTPS. To add a single compute node to an existing control plane instead, use the deltaforge-compute package.

Installed and ready? Browse the live demos or read CSV to Delta Lake Without Spark.

deltaforge-compute

The headless compute node. Deploy as a native binary on any cloud VM via systemd, or as a container in Docker or Kubernetes. The control plane requires a Postgres instance. No JVM, no Spark cluster to manage. Want the entire platform provisioned in one step, compute included? Use Cloud deployment instead.

The standard production deployment. Install the binary (detects x64 or ARM64), drop a systemd unit, and start the service.

install
curl -fsSL https://deltaforge.org/install.sh | sh -s -- --pkg deltaforge-compute
systemd unit
sudo tee /etc/systemd/system/deltaforge-compute.service >/dev/null <<EOF
[Unit]
Description=DeltaForge Compute Node
After=network-online.target
Wants=network-online.target

[Service]
ExecStart=/usr/local/bin/deltaforge-compute
Restart=on-failure
RestartSec=5
Environment=DELTAFORGE_LICENSE_KEY=your-license-key
Environment=DELTAFORGE_CONTROL_PLANE=https://console.deltaforge.org

[Install]
WantedBy=multi-user.target
EOF

sudo systemctl daemon-reload
sudo systemctl enable --now deltaforge-compute

Once running, the node registers itself with your control plane workspace. Add more nodes at any time to scale throughput. See the autoscaling guide.

Useful for local development and testing compute node behaviour before deploying to a cloud VM.

curl -fsSL https://deltaforge.org/install.sh | sh -s -- --pkg deltaforge-compute

Or via Homebrew:

# Add the tap (one-time)
brew tap deltaforge-org/tap
brew install deltaforge-compute

Useful for local development on Windows workstations. For production, use the Linux systemd path.

PowerShell
iwr -useb https://deltaforge.org/install.ps1 | iex; Install-DeltaForge -Pkg deltaforge-compute

Or via winget:

winget install DeltaForge.Compute

How it fits together

Compute nodes run inside your cloud environment and read Delta Lake and Iceberg tables directly from object storage. They report license and metering data to the DeltaForge control plane over outbound HTTPS.

%%{init: {'theme': 'dark', 'themeVariables': {'primaryColor': '#1e1e2e', 'primaryBorderColor': '#b95731', 'lineColor': '#b95731', 'secondaryColor': '#13131a', 'tertiaryColor': '#0d0d14', 'clusterBkg': '#1a1b23', 'clusterBorder': '#444', 'edgeLabelBackground': '#13131a', 'nodeTextColor': '#e0e0e0'}}}%%
flowchart TB
    subgraph clients["Client Tools"]
        pbi["Power BI / Excel / Tableau"]
        db["DBeaver / JDBC Applications"]
        ai["Claude / Cursor via MCP"]
        dfd["DeltaForge Desktop"]
    end

    subgraph cp["Control Plane (self-hosted or cloud)"]
        cpsvc["deltaforge-control-plane :3000"]
        pg[("PostgreSQL")]
        cpsvc --- pg
    end

    subgraph cloud["Your Cloud: AWS, Azure, On-Premises"]
        lb["Load Balancer / ASG / VMSS"]

        subgraph nodes["Compute Fleet - horizontal scale"]
            n1["deltaforge-compute"]
            n2["deltaforge-compute"]
            n3["deltaforge-compute ... +N"]
        end

        subgraph store["Object Storage: S3 / ADLS Gen2"]
            dl["Delta Lake Tables"]
            ice["Iceberg Tables"]
        end
    end

    pbi & db --"ODBC / ADBC"--> cpsvc
    ai --"MCP"--> cpsvc
    dfd --"HTTP API"--> cpsvc
    cpsvc --"query routing"--> lb
    lb --> n1 & n2 & n3
    n1 & n2 & n3 --"Parquet I/O"--> store

AWS EC2

Install on any EC2 instance (x86-64 or ARM64), point table LOCATION paths at s3:// URIs, and IAM role credentials are picked up automatically from the instance profile.

Azure VM

Deploy on any Azure VM or VMSS. Table LOCATION paths use abfss:// (ADLS Gen2) or wasbs:// (Blob Storage) URIs. Managed Identity credentials are picked up automatically.

Container or native binary

Deploy as a Docker container, in Kubernetes via Helm, or as a standalone binary on a VM fleet managed by an Auto Scaling Group or VMSS. The control plane also runs containerised and requires Postgres.

Deploy the platform on your cloud

One template per cloud, written in each cloud's native language: Bicep on Azure, CloudFormation on AWS, Terraform on Google Cloud. A single deployment provisions the whole platform: web console, control plane, managed PostgreSQL catalog, object storage for your tables, a secret store, and HTTPS. Storage auth is keyless (managed identity, IAM task role, or service account), and you do not need a license key up front: the platform provisions a Community license automatically on first boot, attested by your cloud account. Templates live in deltaforge-org/delta-forge/deploy.

DeltaForge on Azure: a Container Apps environment with an always-on platform app and N scale-to-zero compute slots, each direct-dialed on its own managed-TLS FQDN, alongside PostgreSQL flexible server, a storage account, two key vaults and Log Analytics.
What community.bicep provisions on Azure Container Apps. Why these services →

Deploy straight from the portal. The button opens Azure's guided deployment form (resource group, admin accounts, sizing), generated from the same UI definition the Azure Marketplace offer uses:

Deploy to Azure

Or from the command line with the Bicep source. Only the three passwords are required; everything else has sensible defaults:

Azure CLI + Bicep
git clone https://github.com/deltaforge-org/delta-forge
cd delta-forge/deploy/azure

az group create --name deltaforge --location westeurope
az deployment group create --resource-group deltaforge \
  --template-file community.bicep \
  --parameters adminPassword=<min-12-chars> \
               engineerPassword=<min-12-chars> \
               pgAdminPassword=<min-12-chars>

The deployment runs on Azure Container Apps with a pool of scale-to-zero compute slots that the built-in autoscaler starts and parks on demand. Pick the template for your tier: community.bicep (1 slot), standard.bicep (4 slots), or professional.bicep (9 slots). The guiEndpoint output is your console URL; sign in with the admin email and password you chose.

DeltaForge on AWS: a Network Load Balancer fronts an ECS Fargate task running Caddy and the platform, with RDS PostgreSQL in the VPC and S3, Secrets Manager and an IAM task role alongside.
What the CloudFormation stack provisions. Why these services →

One CloudFormation stack provisions everything. Only the three passwords are required; everything else has sensible defaults:

AWS CLI + CloudFormation
git clone https://github.com/deltaforge-org/delta-forge
cd delta-forge/deploy/aws

aws cloudformation deploy \
  --stack-name deltaforge \
  --template-file cloudformation/deltaforge-platform.yaml \
  --capabilities CAPABILITY_IAM \
  --parameter-overrides \
      AdminPassword=<min-12-chars> \
      EngineerPassword=<min-12-chars> \
      PgAdminPassword=<min-12-chars>

# Read the console URL once the stack is up
aws cloudformation describe-stacks --stack-name deltaforge \
  --query "Stacks[0].Outputs[?OutputKey=='ConsoleUrl'].OutputValue" --output text

First boot runs the headless bootstrap: it provisions the catalog schema in RDS, seeds the admin and engineer accounts, and activates the license. Give it a couple of minutes, then browse to the console URL. On stack delete, the S3 tables bucket and a final RDS snapshot are retained so your data survives a teardown.

DeltaForge on GCP: a VPC with a public platform VM and slot_count internal-only compute VMs, the platform being the only public surface, alongside Cloud SQL PostgreSQL, a Cloud Storage bucket, Secret Manager and keyless service accounts.
What terraform apply provisions. Why these services →

Open the Terraform module directly in Cloud Shell, or run it from any machine with Terraform 1.5+:

Open in Cloud Shell
Terraform
git clone https://github.com/deltaforge-org/delta-forge
cd delta-forge/deploy/gcp/terraform

cp terraform.tfvars.example terraform.tfvars
# set project_id; everything else has sensible defaults

terraform init
terraform apply

terraform output console_url

Admin and engineer passwords are auto-generated: read them with terraform output -raw admin_password, or from Secret Manager. First boot installs Docker on the VM, mounts the data disk, and runs the headless bootstrap; give it a few minutes before browsing to console_url.

What each template provisions

The same topology on every cloud, mapped to that cloud's managed services. No access keys are stored anywhere: the engine reaches object storage through the platform's own cloud identity.

  • Azure Container Apps with managed TLS
  • Scale-to-zero compute slot pool, flipped by the built-in autoscaler
  • PostgreSQL Flexible Server catalog
  • Blob Storage for tables (abfss://)
  • Key Vault secrets, managed identity auth
  • ECS Fargate task behind a Network Load Balancer
  • RDS for PostgreSQL catalog
  • S3 bucket for tables (s3://)
  • EFS for durable engine state
  • Secrets Manager, IAM task-role auth
  • Compute Engine VM (Ubuntu + Docker) on a static IP
  • Cloud SQL for PostgreSQL catalog
  • GCS bucket for tables (gs://)
  • Persistent disk for durable engine state
  • Secret Manager, service-account auth

Need a single compute node joining an existing control plane instead? Switch to deltaforge-compute above. For how compute scales up and down after the deploy, see the autoscaling guide.

deltaforge-cli

Single binary for scripting queries, running SQL files, executing pipelines, and checking cluster health from any terminal or CI environment.

curl -fsSL https://deltaforge.org/install.sh | sh -s -- --pkg deltaforge-cli

Or via Homebrew (add the DeltaForge tap once, then install deltaforge-cli as a formula):

# Add the tap (one-time)
brew tap deltaforge-org/tap
brew install deltaforge-cli

Tap source: github.com/deltaforge-org/homebrew-tap

PowerShell
iwr -useb https://deltaforge.org/install.ps1 | iex; Install-DeltaForge -Pkg deltaforge-cli

Or via winget (installs the signed binary into your PATH) or Scoop (the preferred choice for portable, no-admin installs):

winget
winget install DeltaForge.CLI
Scoop 
# Add the bucket (one-time)
scoop bucket add deltaforge https://github.com/deltaforge-org/scoop-bucket
scoop install deltaforge-cli

Detects your architecture (x64 or ARM64), downloads the matching tarball, and drops the binary on your PATH:

curl -fsSL https://deltaforge.org/install.sh | sh -s -- --pkg deltaforge-cli

Or manually:

# Substitute the latest version from the releases page
curl -fsSL https://github.com/deltaforge-org/delta-forge/releases/download/v1.0.0/deltaforge-cli-1.0.0-linux-x64.tar.gz \
  | sudo tar -xz -C /usr/local/bin

Common commands

A few things you can do from the terminal once the binary is on your PATH.

Run a SQL file

Execute any SQL script against a DeltaForge workspace and stream results to your terminal.

deltaforge-cli run query.sql

Execute a pipeline

Trigger a numbered pipeline script and watch step-by-step progress in the terminal.

deltaforge-cli pipeline run --file pipeline.sql

Health check

Check the reachability and version of every registered compute node in a workspace.

deltaforge-cli status

deltaforge-mcp

Plugs DeltaForge into any MCP-compatible AI assistant. Exposes your Delta Lake catalog, pipeline graph, lineage, and query execution as typed tools an AI can call directly. No custom prompts, no clipboard-pasting schemas.

curl -fsSL https://deltaforge.org/install.sh | sh -s -- --pkg deltaforge-mcp

Or via Homebrew (add the DeltaForge tap once, then install deltaforge-mcp as a formula):

# Add the tap (one-time)
brew tap deltaforge-org/tap
brew install deltaforge-mcp

Tap source: github.com/deltaforge-org/homebrew-tap

PowerShell
iwr -useb https://deltaforge.org/install.ps1 | iex; Install-DeltaForge -Pkg deltaforge-mcp

Or via winget (installs the signed binary into your PATH) or Scoop (the preferred choice for portable, no-admin installs):

winget
winget install DeltaForge.Mcp
Scoop 
# Add the bucket (one-time, skip if already added)
scoop bucket add deltaforge https://github.com/deltaforge-org/scoop-bucket
scoop install deltaforge-mcp

Detects your architecture (x64 or ARM64) and downloads the matching binary. --pkg deltaforge-mcp selects the MCP server instead of the default CLI package:

curl -fsSL https://deltaforge.org/install.sh | sh -s -- --pkg deltaforge-mcp

Wire it into your AI assistant

Point your AI assistant's MCP config at the server binary. The format is the same across all MCP-compatible tools.

mcp config
{
  "mcpServers": {
    "deltaforge": {
      "command": "deltaforge-mcp",
      "args": ["--workspace", "your-workspace-id"]
    }
  }
}

Claude Desktop

Add the config block to claude_desktop_config.json: ~/Library/Application Support/Claude/ on macOS or %APPDATA%\Claude\ on Windows. Restart Claude Desktop to pick it up.

Cursor

Open Cursor Settings, navigate to the MCP section, and paste the config block. Cursor reloads MCP servers without restarting, so the tools become available immediately.

VS Code with Copilot

Add the config block to your workspace or user settings.json under the mcp.servers key. The Copilot chat panel picks up the server on the next activation.

Full reference of exposed tools: MCP server documentation.

Got it installed. What now?

Register a workspace in the console, point it at a Git repo with SQL in it, and run your first query.

Open the Console Read the Docs