2026-11-18 18:44:39 -08:00
2026-08-20 20:13:34 -07:00
2026-02-14 05:26:49 -08:00

stack

Hanzo Stack

OSS quickstart for the Hanzo AI infrastructure. One docker compose up -d to spin up the whole platform locally — databases, LLM gateway, chat, search, observability, and supporting services on published images from ghcr.io/hanzoai/*.

git clone https://github.com/hanzoai/stack.git
cd stack
docker compose up -d

For development of Hanzo services themselves, use hanzo/universe (the mono). For production Kubernetes, use hanzo/operator (20-CRD k8s operator).

This repo's root remains the image-only infrastructure quickstart. The optional FanClub AI reference application lives under examples/ and does not alter root Compose or production orchestration.

What's in it

Just compose files + minimal config. Everything is a published image from ghcr.io/hanzoai/*. No source, no submodules, no build step.

Service Image Local port
Postgres ghcr.io/hanzoai/sql 5432
Redis / KV ghcr.io/hanzoai/kv 6379
DocDB ghcr.io/hanzoai/docdb 27017
MinIO minio/minio 9000 / 9001
Meilisearch getmeili/meilisearch 7700
Qdrant qdrant/qdrant 6333
Temporal temporalio/auto-setup 7233
LLM Gateway ghcr.io/hanzoai/llm 4000
Chat ghcr.io/hanzoai/chat 3081
MCP ghcr.io/hanzoai/mcp varies
Agent ghcr.io/hanzoai/agent varies
Runtime ghcr.io/hanzoai/runtime 8002
Search ghcr.io/hanzoai/search 3000
IAM ghcr.io/hanzoai/iam 8000
Cloud ghcr.io/hanzoai/cloud 3001
Commerce ghcr.io/hanzoai/commerce 8001
Console ghcr.io/hanzoai/console 3060
Base ghcr.io/hanzoai/base 8090

See compose.yml for the full list. Per-environment overrides in compose.dev.yml, compose.core.yml, etc.

Optional reference application — FanClub AI

A fictional, industry-neutral organization intelligence prototype is available at examples/fanclub-ai/. It demonstrates persistent synthetic records, three independent tenant workspaces, decision baselines, human-approved simulated actions, audit and optional private Hanzo Engine inference.

cd examples/fanclub-ai
npm start       # Node.js 22.13+, no Docker required
npm test        # synthetic-data / tenant-boundary tests

This is an isolated demonstrator, not a production-integrated Hanzo service. Its demo SQLite and tenant selector must be replaced by authenticated Hanzo Cloud + Framework/Base + IAM; see integration plan. It uses no live customer data and sends no external actions. Core infrastructure remains the published-image quickstart described below.

Configuration

Copy the example env file and edit:

cp .env.example .env

The only values you typically need to set for a quickstart are API keys for the LLM gateway:

OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...

Database credentials, ports, etc. all default sensibly.

Pinning versions

The default compose uses :latest. For anything beyond quickstart, override in your own compose.override.yml:

services:
  llm:
    image: ghcr.io/hanzoai/llm:v1.42.0
  chat:
    image: ghcr.io/hanzoai/chat:v0.7.3

Or fork this repo and pin to whatever semver you want.

Profiles

docker compose up -d                                       # everything
docker compose -f compose.yml -f compose.core.yml up -d    # databases + LLM + chat
docker compose -f compose.yml -f compose.dev.yml up -d     # dev-ergonomic ports + volumes

Going beyond the quickstart

This stack is intentionally simple. If you need:

Need Use instead
Develop / modify a Hanzo service hanzo/universe — the mono with full source
Run in production Kubernetes hanzo/operator — 20-CRD operator (Service, Datastore, SQL/KV/DocDB/S3 facades, Gateway, MPC, Network, Ingress, DNS, more)
Run in production PaaS platform.hanzo.ai — Dokploy-based PaaS, deploys from git
Multi-tenant SQLite + extension runtimes hanzo/base — embedded in every Hanzo Go service
LLM gateway only hanzo/llm — 100+ providers, OpenAI-compatible
Identity (OAuth/OIDC/SAML) hanzo/iam
Observability hanzo/console (traces/scores/prompts) + hanzo/o11y

Troubleshooting

docker compose logs <service-name>      # logs
docker compose restart <service-name>   # restart one service
docker compose down -v                  # nuke everything, including volumes

License

MIT. See LICENSE.

S
Description
Docker Compose quickstart to run the Hanzo AI platform locally.
Readme MIT
29 MiB
Languages
Python 98.9%
Shell 0.6%
Makefile 0.5%