Open-source agent platform

Build AI agentsfor your business.

Launch business-ready agents fast with your data, tools, workflows, models, and skills.

Open sourceSelf-host readyFull control

Agent Runtime

Run business agents with Skills.

ZGI gives agents a runtime to load Skills, use company knowledge and live data, call tools, and return useful work fast.

Active Skills

Renewal RiskCustomer BriefFollow-up Email

AIChat

Business workspace

GPT-5.5

Which renewals need attention this week?

gpt-5.5Working

Renewal Risk loaded<1ms

Knowledge searchedQueued

Customer data queriedQueued

Ask ZGI anything...

Context Layer

Give agents the contextthey need

ZGI connects files, databases, knowledge bases, memory, and previous work so Skills can act with the right business context.

Agents need more than a prompt.

They need trusted context, structured data, unstructured knowledge, memory, and source-aware recall.

Mainline

Sources → Context → Skills → Execution

Structured DataUnstructured Data

05 / Policy

Scope

Keep every retrieval and action inside approved workspace, data, and permission boundaries.

PermissionsSources

Business context

Files & Docs

PDFs, webpages, meeting notes

Databases

Accounts, revenue, metrics, tickets

Memory

Preferences, decisions, past work

ZGI Runtime

Load context into Skills

Search knowledge

Policies, docs, meeting notes

32ms

Query structured data

Accounts, revenue, tickets, renewals

58ms

Attach workspace memory

Preferences, decisions, handoffs

Done

Check permission scope

Approved data and tools only

Safe

Business Skills

Renewal Risk

Find accounts at risk this week

Customer Brief

Create account briefs and next steps

Follow-up Email

Draft a ready-to-send follow-up

Output: 3 at-risk accounts, reasons, owners, and follow-up emails

Source-aware, reviewable, ready for team handoff

Ready

LLM Gateway

Manage everymodel channel

ZGI brings provider catalogs, runtime channels, default models, API keys, quotas, and usage logs into one OpenAI-compatible gateway.

Route through channels

Use priority, weight, availability, and quota to decide where each call runs.

Control API access and usage

Manage API keys, switches, spend, remaining quota, and auditable usage records.

Explore the Gateway

Model Channel Management

Providers define capabilities. Channels run them.

5 enabled channels

Channels

109

Supported models

Usable models

148.5M

Remaining quota

Runtime channels

Priority / Weight

ZGI Cloud

Official

Official · 102 models · 148.5M pts

P100

W100

qwen

Enabled

Alibaba Cloud · 32 models · 99.9M pts

P100

W100

deepseek

Enabled

DeepSeek · 9 models · 996K pts

P100

W100

豆包

Enabled

Volcengine · 9 models · 999K pts

P100

W100

Default capabilities

Chat, embedding, rerank, vision, and image generation

Chat

Qwen3.6-Flash

Embedding

Text Embedding V4

Rerank

GTE Rerank V2

Vision

Doubao-Seed-2.0-Pro

Image Gen

gpt-image-2

OpenAI-compatible API

https://api.zgi.app

/v1/chat/completions

/v1/responses

/v1/embeddings

/v1/models

API key / usage trail

sk-•••••••• · chat.completions · qwen → deepseek fallback

Auditable usage