Building in the
open.
We publish what we build, fix the bugs we find upstream instead of patching them in private, and pay the projects we depend on. The commitments below are specific enough to hold us to.
What Hanzo is built on
The projects our platform actually runs on, each with its licence and the people who maintain it.
The database underlying all our stateful services
PostgreSQL LicenseOur Stewardship Commitments
We hold ourselves to explicit, public commitments — not vague marketing language.
We ship open source first
Every tool we build for ourselves gets published. If it's useful to us, it belongs to the community. No internal-only forks.
We contribute upstream
When we find bugs or missing features in projects we depend on, we fix them upstream — not in private patches.
We never lock in what we open
Anything published under MIT or Apache-2.0 stays that way. We will never relicense open repos to extract commercial value.
We credit our foundations
Our platform is built on hundreds of open-source projects. We acknowledge, link to, and financially support the communities and maintainers who make this possible.
We fund what we depend on
Through GitHub Sponsors, Open Collective, and direct partnerships, we invest back into the projects that power Hanzo.
We welcome all contributors
Good code can come from anywhere. We maintain a contributor-friendly CONTRIBUTING.md, respond to PRs within 48 hours on active repos, and recognize every contributor.
20% to the author of the code that runs
Verify that you own a repository, and every org that deploys a project built from it pays you 20% of its metered spend, accrued each period.
69 organizations checked · non-fork repositories
repositories with a commit in 30 days
- hanzoai/commercefix(billing): surface detailed card decline reasons and handle PAN_FAILURE for token top-ups
- hanzoai/devrelease: bump Dev to 0.7.4
- hanzo-docs/docsfix(docs): remove unsubstantiated Kai 1.9 claims, retain reproducible measurements
- hanzoai/hipsfix(hip-1332): retain only reproducible 12-suite harness measurements in §22
- hanzoai/papersfix(kai): replace unsupported claims with reproducible 12-suite harness metrics
- hanzoai/senkaiProvide pinned curl headers for client checks without system package installation
- hanzoai/docsLocalize the docs framework across 16 languages (#5)
- hanzoai/extensionfeat(extension): support responses API and robust ProseMirror polling for CLI dev loop
Public GitHub snapshot · read Oct 29, 17:40
Start with the foundations.
Selected open projects with direct paths to their source code.
One ecosystem. Many layers.
72 Hanzo-branded organizations, plus related ecosystem projects in one directory.
Enterprise solutions, custom deployments, and client architectures.
Shared application framework, routing, and state hydration.
First-party web applications, consoles, and ecosystem dashboards.
Common Lisp client libraries and symbolic reasoning bindings.
Open source community contributions, hackathons, and guides.
Kernel services and low-level protocol implementations.
Native C++ bindings and low-level system runtimes.
Dart client packages and Flutter widgets.
Official documentation source for platform, SDKs, and APIs.
Showing 9 of 72 matched organizations
1,617 public repositories across 69 organizations, with the source on git.hanzo.ai and mirrored on GitHub.
See what we’re building.
Read the source. Explore the stack. Build something better.
OSS Catalog
Open source tools, templates, and infrastructure for building AI applications
Quick Start Collections
46 results
Hanzo Picks
Hanzo AI
hanzoai/ai
One API for every model. OpenAI- and Anthropic-compatible, with load balancing, caching, and observability.
Hanzo Dev
hanzoai/dev
AI coding agent for your terminal. Multi-agent orchestration, codebase understanding, and PR automation.
Hanzo MCP
hanzoai/mcp
260+ tools for AI models via Model Context Protocol. File ops, browser, search, memory, and more.
Approvals
hanzoai/approvals
Human-in-the-loop approval workflows. Review gates for sensitive AI operations.
Ellipsis
ellipsis/ellipsis
Shell-based dotfiles package manager with extensions ecosystem.
GitHub Connector
hanzoai/github-connector
Sync GitHub repos, issues, and PRs. Bi-directional integration.
Go SDK
hanzo-go/sdk
Full cloud SDK for Go. Idiomatic Go with context support and connection pooling.
Hanzo Bot
hanzoai/bot
Autonomous AI team in a box. Every role, every channel, every platform.
Hanzo Chat
hanzoai/chat
Enhanced ChatGPT-style UI with multi-model support, plugins, and MCP integration.
Kotlin SDK
hanzo-kt/sdk
Full cloud SDK for Kotlin. Coroutine-native client for JVM and Android, installed from JitPack.
KV Store
hanzoai/kv
Edge-optimized key-value store. Sub-millisecond reads with global replication.
Policy Engine
hanzoai/authz
Policy-as-code for AI governance. Define rules for model access, data handling, and compliance.
Python SDK
hanzoai/python-sdk
Flagship Python SDK — models, agents, tools, memory, and the full Hanzo cloud. pip install hanzo.
Slack Connector
hanzoai/slack-connector
Slack integration for AI workflows. Commands, notifications, and threads.
SQL
hanzoai/sql
Managed relational database with pgvector. SQL and vector search in one store.
Swift SDK
hanzo-swift/sdk
Full cloud SDK for Swift. Idiomatic async/await SwiftPM package for iOS, macOS, and server-side Swift.
TypeScript SDK
hanzo-js/sdk
Full cloud SDK for TypeScript/JavaScript. Works in Node.js, browsers, and edge runtimes.
Zen Nano
zenlm/zen-nano
0.6B edge model — 40K context, optimized for on-device deployment.
Zen VL 30B
zenlm/zen-vl-30b-instruct
30B MoE vision-language model — STEM/code multimodal reasoning.
Zen VL 4B
zenlm/zen-vl-4b-instruct
4B vision-language model — image analysis, OCR, multimodal reasoning.
Zen VL 8B
zenlm/zen-vl-8b-instruct
8B vision-language model — 256K context, OCR across 32 languages.
Zen4 Coder
zenlm/zen4-coder
80B MoE code model — abliterated open weights for generation, review and debugging.
Zen4 Coder Flash
zenlm/zen4-coder-flash
31B MoE fast code model — 131K context, 59.2% SWE-bench, tool calling.
Zen4 Max
zenlm/zen4-max
30B MoE unbiased AI model — 3B active parameters, efficient inference.
Zen4 Ultra
zenlm/zen4-ultra
1.04T MoE frontier model — 32B active parameters, 256K context.
Zen4 Ultra Trainer
zenlm/zen4-ultra-trainer
GT-QLoRA training for 1T MoE models — Gate-Targeted QLoRA for MoE uncensoring.
Contribute & Earn
We welcome contributions of all kinds — code, documentation, bug reports, and feature requests. Every contribution earns a share of compute revenue through our OSS Revenue Sharing program.