Skip to content
@hanzoai

Hanzo AI

AI infrastructure for the agentic era — LLM gateway, MCP tools, multi-agent SDKs, and dev tools. Techstars '17.

Hanzo

Run the AI cloud on your own machine.

Everything below is open source. The same code runs our cloud and yours.

hanzo.ai · docs.hanzo.ai · cloud.hanzo.ai


Start here — the local stack

Most AI platforms give you an API key. We give you the platform. One binary, and hanzo serve brings up the cloud itself on your own hardware.

curl -fsSL https://hanzo.sh | sh     # the CLI, the MCP server, the coding agent
hanzo                                # an AI engineer in your terminal

hanzo serve cloud|iam|kms|gateway|storage|pubsub   # the cloud, on your machine
hanzo engine serve <model>                          # an OpenAI-compatible endpoint, local

hanzo-node is not a second program — the installer writes it as a symlink to the same binary. engine serve needs the engine on PATH; it is a separate install and the command tells you so.

Prefer a window to a prompt? Download Hanzo Desktop — it ships the agent node inside the app and talks to the engine for inference.

What it is
desktop The whole stack as a desktop app. The agent node ships inside the bundle as a sidecar; inference comes from the engine, local or on your LAN. macOS · Windows · Linux.
cli One static Rust binary: an AI coding agent, and every product of the Hanzo cloud from the terminal. No runtime, no daemon.
engine The inference engine. Fast, flexible LLM and embedding serving in Rust — the thing that actually runs the model.
ml The compute core underneath it: multi-backend tensors for Rust across CPU · CUDA · Metal · ROCm · Vulkan, quantization built in.
net Pool the machines you already own — iPhone, Mac, Raspberry Pi, NVIDIA boxes — into one inference cluster.
edge On-device inference for mobile, web, and embedded, built on Hanzo ML.

Local and cloud are the same API (/v1/). Point a client at localhost or at api.hanzo.ai and nothing else changes.


Build with AI

What it is
python-sdk The most complete SDK we ship: agents, MCP tools, memory, inference, cloud.
js-sdk TypeScript and JavaScript — agents, cloud, inference.
java-sdk · sdk Typed Java client, and the index of every SDK in every language.
mcp Model Context Protocol server — 260+ tools for agents.
agent Multi-agent framework with an OpenAI-compatible API.
dev A fast local coding agent for your terminal.
code An open source AI code editor — any model, your data stays yours.
extension The same agent inside VS Code and compatible editors.
app Describe an app, get an app. AI web and app builder.
studio Visual AI engine — build pipelines by wiring them up.
chat · bot Chat with MCP and any provider; the same assistant on WhatsApp, Telegram, Slack, Discord.
crawl Web crawling that returns LLM-ready text.

Run the cloud

These are the services behind api.hanzo.ai. Each is a single binary you can run yourself.

What it is
platform The PaaS. Git push to deploy, on your own Kubernetes.
console The admin and observability console for all of it.
iam Identity — OIDC, JWT, per-brand SSO. Never roll your own auth.
gateway The trust boundary: identity, rate limiting, metering, CORS.
authz · hsm Access control for Go (ACL/RBAC/ABAC); HSM signers including ML-DSA.
functions Serverless compute for event-driven work.
tasks Durable workflow execution for agent orchestration.
commerce · finance Orders, subscriptions, metering, credit ledger; a double-entry financial core.
flags · insights Feature flags and A/B; product analytics.
o11y · otel-collector OpenTelemetry-native metrics, traces, logs.

Data

What it is
base The storage substrate: multi-tenant SQLite with in-process polyglot extensions and encrypted replication.
datastore Column-oriented OLAP for real-time analytics.
s3 · vfs S3-compatible object storage; an S3-backed virtual block filesystem.
sql · kv · docdb Postgres with pgvector; a Redis-compatible key-value store; a MongoDB-compatible document database.
vector · index Vector search for embeddings; full-text, vector, and hybrid retrieval in Rust.
orm · sqlite · sqlcipher Type-safe generics ORM for Go; encrypted SQLite at rest.
pubsub · kafka Event streaming, and a Kafka wire-protocol gateway for it.

Infrastructure

What it is
ingress Kubernetes-native L7 proxy and load balancer with automatic TLS.
dns Programmable DNS — CoreDNS with Hanzo plugins.
git Self-hosted Git, code review, package registry, and CI in one service.
ci · cd · pack Build, test, and deploy any repo from one hanzo.yml; declarative CD for Kubernetes; zero-config BuildKit builds.
registry Container registry with Hanzo IAM token auth.
mail Self-hosted mail — SMTP, IMAP, webmail, DKIM/SPF/DMARC.

Applications

Complete products, open source, running on the stack above.

cms · erp · helpdesk · esign · captable · dataroom · cal · social · world · tabs · frames

Design

ui · gui · shadcn · svelte · design · brand · logo · gallery

Go libraries

Small, single-purpose, no framework attached.

money · decimal · doctype · framework · migrate · notify · research


Read the papers · Ship on cloud.hanzo.ai

Hanzo AI, Inc. — Techstars '17

Popular repositories Loading

  1. ai ai Public

    Hanzo AI — LLM control plane, RAG, and model hub. Native Go routing for 66+ models over an OpenAI-compatible /v1 API; mounts as the 'ai' subsystem inside hanzoai/cloud.

    Go 3

  2. app app Public

    Hanzo App — AI web & app builder on Hanzo Cloud (MIT, derived from DeepSite)

    TypeScript 3 1

  3. cuckoo cuckoo Public

    🐦 Event recording library.

    CoffeeScript 1

  4. form.js form.js Public

    🖍️ Automagic forms.

    CoffeeScript 1

  5. ui ui Public

    React component library for AI applications

    TypeScript 1

  6. platform platform Public

    Hanzo Platform — unified PaaS for deploying AI applications

    TypeScript 1

Repositories

Showing 10 of 281 repositories

Top languages

Loading…

Most used topics

Loading…