Skip to content
DashboardGPT

Self-hosted deployment

Conversational analytics that runs inside your network.

Deploy DashboardGPT on your own servers with Docker or Kubernetes. Connect your own LLM key, or run a model locally and stay fully air-gapped. Your data, queries and prompts never leave your infrastructure.

For the complete AI-readable content index, see /llms.txt.

System requirements
Install with Helm (license key from your account team)$ helm repo add dashboardgpt <private-registry>$ helm install dgpt dashboardgpt/dashboardgpt \    --namespace analytics --create-namespace \    --set license.key=$DGPT_LICENSE \    --set llm.provider=ollama✓ app      3/3 ready✓ workers  4/4 ready✓ https://analytics.your-company.internal
  • Built on Apache Superset
  • SSO: SAML · OIDC · LDAP
  • Row-level security
  • Audit logs
  • Local LLM ready

Deployment options

Three ways to run it on your infrastructure

Same product as the cloud version. Pick the footprint that fits your team and your security policy.

  • Single host

    Docker Compose

    One VM or bare-metal server runs the app, workers, PostgreSQL and Redis.

    • Fastest path to a working install
    • Bundled or external PostgreSQL
    • Backups with standard pg_dump

    Best for: pilots, departments and teams up to about 50 users.

  • High availability

    Kubernetes with Helm

    Production chart for EKS, AKS, GKE, OpenShift or your own cluster.

    • Scale app pods and query workers separately
    • Use your managed PostgreSQL and Redis
    • Rolling upgrades with helm upgrade

    Best for: company-wide rollouts and platform teams.

  • Zero egress

    Air-gapped with a local LLM

    Offline bundle and license file. The language model runs on your own GPUs.

    • Works with Ollama, vLLM or any OpenAI-compatible endpoint
    • No outbound network calls
    • Signed images with checksums

    Best for: government, defence, banking and healthcare.

Architecture

Everything inside the dashed line runs on your servers

DashboardGPT sends SQL to your databases, gets aggregated results back and asks the model to write the chart and the insight. With a local model, that whole loop stays inside your network.

On-premise architectureUsers reach DashboardGPT through your load balancer. The app, workers, PostgreSQL metadata database and Redis run inside your network and query your databases. The LLM gateway calls either a local model inside the network or, optionally, your own cloud LLM provider.your-network.internalYour usersSSO · browserLoad balancerTLS · ingressDashboardGPT appcanvas · chat · insightsSuperset engineLLM gatewayaudit logWorkersCeleryRediscache · queuePostgreSQLmetadata · canvasesYour databasesPostgres · SQL ServerOracle · SnowflakeBigQuery · ClickHouse …Local LLMOllama · vLLM · GPUYour LLM keyoptional · BYOKegress optional
SQL to your dataModel callsOptional cloud LLM (BYOK)

Users reach DashboardGPT through your load balancer. The app, workers, PostgreSQL metadata database and Redis run inside your network and query your databases. The LLM gateway calls either a local model inside the network or, optionally, your own cloud LLM provider.

Cloud or self-hosted

Choose where DashboardGPT runs

The same workspace either way. What changes is who operates it and where the data goes.

CloudOn-premise
Where it runsOur managed infrastructureYour data centre or your own cloud account
Data locationResults processed in our cloud over TLSData, prompts and results stay in your network
Language modelYour own key (BYOK)Your own key, or a local model with no egress
Sign-inEmail, Google, MicrosoftSAML, OIDC, LDAP / Active Directory
UpgradesAutomaticYou choose when, with versioned releases
Who operates itDashboardGPTYour IT team, with our support
Time to startMinutes, free plan availableAfter a scoping call and trial license
Best forTeams that want to start todayRegulated data and strict security policies

Not sure yet? Start on the cloud while your security review runs, and move to self-hosted when it clears. We help migrate your workspaces and canvases. Start on the cloud.

Security and governance

Controls your security team will ask about

The hosted controls on the security page still apply. Running it yourself adds the ones below.

  • Single sign-on

    Connect SAML 2.0 or OpenID Connect identity providers such as Okta, Entra ID and Keycloak, or LDAP / Active Directory directly.

  • Roles and row-level security

    Control who sees which databases, tables and rows. A regional manager asking a question only gets answers from their region's data.

  • Audit logs

    Every question, generated query and data access is logged with the user and time, ready to ship to your SIEM.

  • Your keys, encrypted

    Database credentials and LLM keys are encrypted at rest with a secret you control, scoped per workspace.

  • Network control

    Run with no outbound access, or allow egress only to the LLM provider you choose. Nothing calls home with your data.

  • Read-only by design

    Connect with read-only database users. DashboardGPT queries your data and stores only canvases and metadata in its own PostgreSQL.

Data sources

Connects to the databases you already run

CSV and Excel uploads work today. Database and warehouse connections are coming soon, for anything Apache Superset can query, including:

  • PostgreSQL
  • MySQL / MariaDB
  • Microsoft SQL Server
  • Oracle
  • Snowflake
  • BigQuery
  • Amazon Redshift
  • Databricks
  • ClickHouse
  • Trino / Presto
  • Amazon Athena
  • SAP HANA
  • DuckDB
  • CSV / Excel

Rollout

From first call to live in your environment

Most teams are running a pilot within two weeks of the first call.

  1. 01

    Scoping call

    We review your users, data sources, LLM choice and security requirements.

  2. 02

    Trial license

    You get images, the Helm chart or Compose bundle and a time-limited key.

  3. 03

    Install

    Your team installs from our docs, with a guided session from our engineers.

  4. 04

    Connect

    Add read-only database users, SSO and your model endpoint.

  5. 05

    Go live

    Roll out to teams with roles, row-level rules and analytics templates.

Questions

On-premise questions

Deployment

You get access to signed container images and a Helm chart from our private registry, plus a Docker Compose bundle for single-host installs. Air-gapped customers receive an offline bundle (images, chart and checksums) to load into their own registry.

Only if you want to run the language model locally. With bring-your-own-key, DashboardGPT calls your existing OpenAI, Anthropic, Azure OpenAI or Google provider, or any OpenAI-compatible endpoint, and the application itself runs on standard CPU servers. See the system requirements page for GPU sizing.

Any provider supported by our bring-your-own-key setup, or any self-hosted model exposed through an OpenAI-compatible API, such as models served by Ollama or vLLM. You choose the model per workspace.

No customer data, queries, prompts or results leave your network. Connected installs make a license check; air-gapped installs use an offline license file and make no outbound calls at all.

We publish versioned releases with release notes. On Kubernetes you upgrade with helm upgrade; on Docker Compose you pull the new images and restart. Database migrations run automatically, and you can stage an upgrade in a test environment first.

Your IT or platform team can install it from our documentation. Every license includes a guided installation session with our engineers, and we can do the install with you over a screen share.

Yes. Start on the hosted version while security review runs, then move to self-hosted. We help you move workspaces, data connections and canvases across.

We don't hold SOC 2 yet. We complete your security questionnaire, and the on-premise edition keeps all data inside your network.

DashboardGPT runs on Apache Superset with a PostgreSQL metadata database and Redis for caching and background jobs. That means it connects to the same wide range of SQL databases Superset supports.

See DashboardGPT running on infrastructure like yours.

Tell us about your data and security setup. We will walk through the architecture, answer your security questionnaire and set you up with a trial license.

  • 30-minute call with a solutions engineer
  • Architecture and sizing review
  • Trial license for your own servers
Contact sales