ZGI

English · 简体中文 · 日本語 · 한국어

A source-available Agent Runtime workspace for building, connecting, publishing, and operating AI agents and executable workflows.

_{Why ZGI · How it works · Capabilities · Quick Start · Development · Docs · Contributing · License}

Why ZGI

ZGI is an Agent Runtime platform with source code available for teams that need AI applications to execute real work, not just answer in a chat box. It brings agent configuration, visual workflow orchestration, advanced retrieval, structured data, model routing, reusable skills, and sandboxed execution into one self-hostable workspace.

Build an agent once, bind it to approved knowledge, databases, skills, and workflows, then make it available through a WebApp, the internal app center, an API, or scheduled and internal calls. Permissions, runtime logs, and batch tests keep the application observable and governable after it is published.

From Build to Runtime

Build agents and workflows
        ↓
Connect models, knowledge, databases, files, and skills
        ↓
Execute tools, code, retrieval, and human-in-the-loop steps
        ↓
Publish through WebApp, App Center, API, or internal invocation
        ↓
Operate with permissions, logs, and batch testing

Core Capabilities

Area	What ZGI provides
Agent applications	Configure instructions, models, memory, knowledge, file inputs, skills, and workflow bindings, then publish a usable agent experience.
Executable workflows	Orchestrate LLM calls, branches, loops, approvals, user questions, HTTP requests, database operations, code, documents, notifications, and scheduled tasks on a visual canvas.
Advanced retrieval	Combine semantic, full-text, hybrid, and knowledge-graph retrieval with reranking, while keeping agents scoped to approved knowledge and data.
Skills and sandboxed tools	Package reusable capabilities for files, charts, reports, calculations, databases, and workflow calls, and execute them in an isolated runtime.
Model gateway	Manage providers, channels, credentials, defaults, routing policies, quotas, and pricing metadata in one place.
Publishing and governance	Expose agents through WebApp, App Center, API keys, or internal calls, with workspace permissions, runtime logs, and reusable batch tests.
Self-hosted runtime	Run the console, API, sandbox, runner, PostgreSQL, and Redis locally or in your own infrastructure.

Quick Start

Start the full local stack:

make dev-docker

If you do not have make, run the startup script directly:

./dev/start-docker

Open the console:

http://localhost:2679

On first launch, create the first administrator account. ZGI does not ship with a default admin account.

Stop the stack:

make docker-down

View logs:

make docker-logs

Development

For source development, install:

• Docker and Docker Compose
• Make
• Go
• Node.js and pnpm

The web app uses pnpm@10.12.1.

Prepare dependencies:

make setup

Run the API and web app from source in separate terminals:

make dev-docker
make dev-api
make dev-web

Documentation

Read the product documentation at docs.zgi.ai.

Repository-local README files are kept for development and contribution notes. For deployment behavior such as the embedded system skill catalog, see docker/README.md.

Contributing

Contributions are welcome. Please read CONTRIBUTING.md before opening a pull request.

Community expectations are documented in CODE_OF_CONDUCT.md.

For security-sensitive reports, follow SECURITY.md.

License

ZGI source code is available under the ZGI Community License, based on Apache License 2.0 with additional conditions. ZGI is free for personal, research, educational, and internal organizational use. Hosted multi-tenant services, white-label distribution, and removal of official ZGI branding require a commercial license. This license is not an OSI-approved open source license. See LICENSE for details.

The Apache License 2.0 text referenced by the ZGI Community License is included in LICENSE-APACHE.