Self-hosted

Self-Hosted AI Database Gateway — Docker, Compose, Kubernetes

Q: Can I use my own LLM gateway?

Yes. FUTRIXDATA_AI_BASE_URL can point at any OpenAI-compatible endpoint — internal gateway, Ollama, vLLM, TGI, or a managed provider routed through your network.

FutrixData Enterprise Edition runs inside your network as a long-lived container. Database credentials never leave your perimeter. A single Docker image, optional external PostgreSQL backend, runs as non-root, and integrates with private model gateways so the full agent-to-data path stays under your control.

Download for macOS Read the docs

Why self-host an AI database gateway

Data residency. Database credentials and query results stay inside your network perimeter.
Sovereign clouds and regulated environments. Financial services, healthcare, public sector — anywhere "vendor sees production data" is not an option.
Centralized governance. One container is the agent-to-data layer for the whole org, instead of every developer wiring up their own connections.
Instant revocation. Per-agent keys are managed in your console; you don't depend on a vendor support ticket to cut off access.

Deploy options

Docker. Single multi-stage image. Slim runtime layer.
Docker Compose. Bundled compose file exposes ports, data directories, and AI settings as environment variables. docker compose up -d --build on any host with Docker.
Kubernetes. Same image, deployed via Helm chart or your own manifests. Works on any container platform — Kubernetes, Nomad, internal servers.

What stays inside your control boundary

Database credentials (no SaaS callback ever holds them).
Query bodies, parameters, and result rows.
Audit log records.
Sensitivity classification reports.
AI inference, when pointed at an internal model gateway or private model deployment.

Architecture

Multi-stage Dockerfile: build UI, compile backend, produce slim runtime.
Persistent volumes hold data sources, AI configuration, risk rules, sensitivity policies, and history across container replacements.
Optional external PostgreSQL backend stores runtime state for centralized backup, monitoring, and HA.
Containers run as a non-root user by default.

Private model gateway support

If your security policy requires AI inference to stay on-premises, point FutrixData's AI service endpoint at an internal model gateway (BYOK with Anthropic / OpenAI / Google routed via your gateway, or local models via Ollama / vLLM / TGI). The full agent-to-data path then remains under your control: no third-party LLM provider, no SaaS proxy, no vendor in the path.

Examples

One-command internal deployment

docker compose up -d --build

Compose snippet

services:
  futrixdata:
    image: futrixdata/enterprise:latest
    ports:
      - "8080:8080"
    environment:
      FUTRIXDATA_DB_URL: "postgres://fxd:***@db:5432/futrixdata"
      FUTRIXDATA_AI_BASE_URL: "https://internal-llm-gateway.example.com/v1"
      FUTRIXDATA_AI_MODEL: "claude-sonnet-4-6"
    volumes:
      - fxd-data:/data
    user: "1001:1001"

volumes:
  fxd-data:

Frequently asked questions

Can FutrixData run air-gapped?

Yes — provided your internal model gateway can serve completions for the built-in AI features (or you turn the built-in AI off). The proxy itself, risk engine, classification store, and audit log all run locally without outbound calls.

What hardware do I need?

A single small VM (2 vCPU / 4 GB RAM) handles a workgroup. Production deployments typically run 2–4 vCPU per replica with the audit log on external PostgreSQL for HA.

Where is state persisted?

By default, in a Docker volume. For production, point FutrixData at an external PostgreSQL — that gives you the same backup, restore, and monitoring story you already use for the rest of your stack.

How do upgrades work?

Replace the image. State lives in volumes or the external database; the container is stateless. Upgrades are usually a Compose / Helm rollout away.

Is there a license?

Enterprise Edition is licensed per deployment. The free desktop app is unrestricted for individual developers; Enterprise unlocks self-hosted server, multi-tenant admin, and external PostgreSQL backend.

Can I use my own LLM gateway?

Yes. FUTRIXDATA_AI_BASE_URL can point at any OpenAI-compatible endpoint — internal gateway, Ollama, vLLM, TGI, or a managed provider routed through your network.

Try FutrixData on your own database

Free desktop app on macOS, Windows, and Linux. Self-hosted Enterprise Edition for production deployments.

Download the desktop app Read the docs