The AI cloud
you can run yourself.
Not another AI API. Hanzo is a decentralized cloud powered by hanzo.network — hanzod nodes spawn the same unified cloud binary we run in production. 67 capabilities behind one API. Open source, one binary. Bring your own GPU or Kubernetes, or run it all on your laptop. Connect any provider, watch usage and cost in one place, and let the native router serve the most optimal model — saving up to 90% on compute.
’17One console for the whole cloud — models, agents, vector search, gateways, IAM, and audit. Run it in our cloud, on your own infrastructure, or on the network, with a complete audit trail.
Backed by builders. Used by teams shipping AI.
Hanzo is backed by leading technology partners and used across AI, commerce, media, infrastructure, and developer tooling.
67 capabilities. One API.
Everything the cloud does is one route — api.hanzo.ai/v1/<capability>. Sixty-seven capabilities, grouped into eight categories. Breadth without the overwhelm.
Identity & Trust
Who you are, what you may touch, where secrets live — IAM, authorization, KMS, and zero-trust.
Intelligence
The mind of the cloud — models, ML, the inference engine, agents, evals, and code execution.
Data
The stores every capability reads and writes — object storage, Base, SQL, KV, and vector search.
Streams
Messaging, durable tasks, and async orchestration — pub/sub, queues, streams, and workflows.
Observability
See everything — telemetry, tracing, and usage analytics across every request.
Commerce
The economy — meter, price, bill, and reward, with billing, plans, and wallets built in.
Platform
The cloud fabric — deploy, provision, route, and host through the native PaaS and gateway.
Applications
User-facing surfaces on top of it all — console, chat, search, git, teams, and knowledge.
Why Hanzo
AI agents need infrastructure, not just prompts.
A good demo can autocomplete a function. A production agent needs context, permissions, tools, memory, tests, secrets, sandboxes, and a way for humans to approve what matters. Hanzo gives agents the operating environment they need to do real work safely.
Policy by default
Permissions, approvals, and rate limits are enforced at the platform — not bolted on per agent.
Evidence over vibes
Every agent run produces a diff, a log, and a test result. Decisions are reviewable, not guessed.
Persistent execution
Long-running sandboxes install dependencies, refactor large codebases, and iterate until tests pass.
Human control
Plan-only, implement, test, deploy — pick the autonomy level. Humans approve what matters.
Race agents. Ship the winner.
Hanzo can run multiple models or agents against the same task in parallel, then compare results using tests, code quality, performance, and policy checks. GPT, Claude, Gemini, Mistral, Groq, xAI, DeepSeek, Qwen, Hugging Face, OpenRouter, Replicate, Together, Fireworks, and local models can all participate in the same workflow.
Parallel attempts
Run multiple models or agents on the same task simultaneously in isolated sandboxes.
Automatic scoring
Score every attempt on tests, code quality, performance, and policy checks.
Best patch wins
Promote the winning diff; archive every attempt with full audit trail.
Works where developers already work.
Hanzo lives in your terminal, your IDE, your browser, and your API — sharing one brain across them all.
Terminal
Install with curl. Drive Hanzo Dev from the shell. Pipe agents into your build.
VS Code & JetBrains
First-class extension surface for editing, refactoring, and reviewing agent diffs.
Browser
Chrome, Firefox, and Safari extensions for in-context agent assistance.
API
OpenAI-compatible HTTP API for models, tools, memory, and agent execution.
Hanzo Dev
AI coding agent in your terminal.
Plan, edit across files, run tests, and open pull requests — driven by your repo, not by guesswork.
- Build APIs, scripts, components, and internal tools
- Fix bugs with full codebase context
- Add tests and CI workflows
- Refactor large systems safely
- Generate deployable services
- Modernize legacy code
One shared brain for your AI tools.
Hanzo connects your models, tools, memory, and history across every surface. Your terminal, IDE, browser, agents, and API calls all share the same context layer.
Shared memory
Context persists across every surface and every session.
Vector search
Semantic search over your entire codebase and knowledge base.
Tool registry
260+ MCP tools available to every agent, IDE, and API call.
Unified history
Every prompt, every response, every tool call — searchable and replayable.
Team sync
Share context, memory, and projects across your team with org-scoped access.
Secure remote execution for teams.
Run long, parallel agent workloads on Hanzo Cloud or on your own Kubernetes — with isolation, quotas, and audit on by default.
Parallel runs
Execute work across repos and tasks simultaneously, with quotas per team.
Shared workspaces
Collaborate via shared sandboxed environments and shared memory.
Quality gates
Enforce policy, tests, and reviews before any change lands on main.
Isolation
Each run gets its own sandbox. Network, file, and tool egress are scoped.
Hanzo API
One API for models, tools, and agents.
Hanzo provides a unified API for chat, model routing, tool calling, memory, and agent execution. It is compatible with existing OpenAI-style SDKs. Change the base URL and start routing through Hanzo.
View API docsimport { Hanzo } from "@hanzo/ai";
const hanzo = new Hanzo({
apiKey: process.env.HANZO_API_KEY,
});
const response = await hanzo.chat.completions.create({
model: "claude-sonnet-4-6",
messages: [{ role: "user", content: "Hello!" }],
});
console.log(response.choices[0].message.content);Hanzo AI, native in every tool you use.
One API key, one identity. Sign in once with @hanzo/iam and the same models, tools, and memory follow you across your browser, editor, docs, and apps — all routed through the @hanzo/ai gateway.
Browser
IDEs
Docs & Office
Design
Comms
Business & CRM
Developer
Desktop & AI clients
Industry
Connect every AI account. One usage plane.
Hanzo is the connective layer for AI — link your accounts once and see limits, credits, and spend everywhere Hanzo runs.
Connect once
Link OpenAI, Anthropic, Hanzo, and 50+ AI providers — plus GPU, cloud, and integration accounts — using logins you already have. No passwords stored.
See unified usage
Rate limits, resets, credits, and spend for every account, in one view — native in the dev CLI, Hanzo Desktop, hanzo.app, hanzo.chat, and console.hanzo.ai.
Share with your org
Add your team to share verified logins and route usage through Hanzo for org-wide analytics and per-member reporting. 1% fee on routed usage.
Every provider. One cloud.
Connect Claude, Gemini, Grok, Groq, DeepSeek, Bedrock — 57+ AI providers in all. Hanzo is the login and usage manager in front of every one: see limits, credits, and spend in one plane, then let the native router serve the most optimal model for each request.
One model id. The right model every time.
Hanzo Inference is the native router: it uses consensus to serve the most optimal model for each prompt — local zen, a cheap tier, or a frontier API — matched to the task and your SLO. Frontier quality without frontier bills: up to 90% lower AI spend, and up to 10× more efficient than a single-provider API.
Send model: "auto"
Point any OpenAI-compatible client at api.hanzo.ai /v1/chat/completions and set model:"auto". One model id — no per-call model picking, no vendor lock-in.
Routed by task + SLO
A built-in heuristic — upgradeable to the trained zen-router — sends each prompt to a local zen model, a cheap tier, or a frontier model, by difficulty and your latency/quality SLO.
Billed as what served you
You pay for the model that actually answered — not a flat premium. Every response carries an X-Routed-Model header, so routing is fully transparent and auditable.
Illustrative, not a guarantee — savings depend on your prompt mix. On the public RouterBench data, oracle routing reaches 0.909 quality at roughly 13× lower cost than the best single model (replication smoke run). Read the research.
Open source, one binary.
The whole cloud is a single static binary you can read, fork, and run. curl -fsSL hanzo.sh | bash and self-host it on your own GPU or Kubernetes — the same cloud we run in production, no lock-in.
Core projects
The complete AI platform for agents, tools, models, and deployment.
260+ Model Context Protocol tools and servers.
Unified gateway for model routing through an OpenAI-compatible API.
AI coding agent for terminal and IDE workflows.
Multi-agent orchestration framework.
Evaluation harness for models, agents, and workflows.
React components for AI applications.
We don't just ship the platform — we research and maintain the Zen family, top open-weight models anyone can run, on-device or in the cloud.
The network is the cloud.
hanzo.network is a decentralized network of hanzod nodes that spawn and power the same unified cloud binary we run in production. Bring any device, mine at market price, and run the whole cloud yourself — no single operator owns it.
Same binary, everywhere
hanzod nodes spawn the exact unified cloud binary we run in production. The network is the substrate; the cloud is one binary on top.
Mine on any device
Bring your own GPU or Kubernetes and earn at market price. Confidential workloads run inside NVIDIA TEE / Confidential Compute.
Free for public + OSS
Public and open-source workloads run free on public nodes. GitHub-for-compute — no gatekeeping, no lock-in.
Pricing
Start free. Scale on your terms.
No credit card to get started. Move to Enterprise when you need higher limits or private deployment.
$0/month
Everything you need to build and ship your first AI agents.
- $5 free credit — no card required
- 60 requests/min, 100K tokens/min
- OpenAI-compatible API
- Chat, agents, and MCP tools
- Community support
- Pay-as-you-go after credit
No credit card required
Custom
For teams that need higher limits, private deployment, SLAs, dedicated support, compliance, and custom infrastructure.
- Higher API and storage limits
- Private cloud or on-prem deployment
- SLAs and dedicated support
- SOC 2 readiness, HIPAA BAA available
- Custom contracts and procurement
- Solution architect engagement
The AI cloud you can run yourself.
Build on it in our cloud, add compute to the network and earn, or self-host the whole thing. One binary, three paths — pick yours.