ZGI gives business AI agents one runtime to use Skills, company knowledge, live data, workflows, models, memory, and tools.
Replies
Best
Hunter
📌
Hey Product Hunt 👋
I’m Eddie from the ZGI team.
We started building ZGI after running into the same problem again and again: building an AI agent demo is relatively easy, but operating agents for real teams is much harder.
Once an agent needs company knowledge, live business data, reusable capabilities, multiple models, workflow execution, permissions, logs, and self-hosting, teams often end up stitching together several different tools and maintaining a lot of glue code.
We wanted to put that runtime layer in one place.
ZGI is a self-hostable Agent Runtime workspace where teams can:
🧩 Give agents reusable Skills for files, charts, reports, calculations, databases, scheduling, and workflow calls
🔀 Build visual workflows with model calls, retrieval, branches, loops, approvals, code execution, and notifications
🧠 Connect agents to approved company knowledge, memory, and live business data
🔌 Manage model providers, credentials, defaults, routing policies, and usage through one gateway
🛠 Run tools inside a sandboxed execution environment
🏠 Deploy the full runtime on your own infrastructure
One area we care about deeply is Skills.
A prompt can tell an agent what to do, but it does not give the agent a reliable way to perform the work. We want teams to turn repeated business capabilities—such as generating a report, querying approved data, creating a file, or calling an internal workflow—into reusable Skills that can be shared across agents.
We’ve made the ZGI source code available because we want developers to inspect the architecture, run it locally, challenge our abstractions, and help us improve it.
For this launch, we would especially appreciate feedback on three things:
1. Where should the boundary sit between a Tool, a Skill, and a Workflow?
2. What runtime and debugging information do you need before trusting an agent in production?
3. What usually makes self-hosting an AI platform painful for your team?
We’re still building, and we know there are many things to improve. Honest feedback, bug reports, issues, and technical discussions would be genuinely useful.
Replies
Hey Product Hunt 👋
I’m Eddie from the ZGI team.
We started building ZGI after running into the same problem again and again: building an AI agent demo is relatively easy, but operating agents for real teams is much harder.
Once an agent needs company knowledge, live business data, reusable capabilities, multiple models, workflow execution, permissions, logs, and self-hosting, teams often end up stitching together several different tools and maintaining a lot of glue code.
We wanted to put that runtime layer in one place.
ZGI is a self-hostable Agent Runtime workspace where teams can:
🧩 Give agents reusable Skills for files, charts, reports, calculations, databases, scheduling, and workflow calls
🔀 Build visual workflows with model calls, retrieval, branches, loops, approvals, code execution, and notifications
🧠 Connect agents to approved company knowledge, memory, and live business data
🔌 Manage model providers, credentials, defaults, routing policies, and usage through one gateway
🛠 Run tools inside a sandboxed execution environment
🏠 Deploy the full runtime on your own infrastructure
One area we care about deeply is Skills.
A prompt can tell an agent what to do, but it does not give the agent a reliable way to perform the work. We want teams to turn repeated business capabilities—such as generating a report, querying approved data, creating a file, or calling an internal workflow—into reusable Skills that can be shared across agents.
We’ve made the ZGI source code available because we want developers to inspect the architecture, run it locally, challenge our abstractions, and help us improve it.
For this launch, we would especially appreciate feedback on three things:
1. Where should the boundary sit between a Tool, a Skill, and a Workflow?
2. What runtime and debugging information do you need before trusting an agent in production?
3. What usually makes self-hosting an AI platform painful for your team?
We’re still building, and we know there are many things to improve. Honest feedback, bug reports, issues, and technical discussions would be genuinely useful.
Thanks for taking a look at ZGI.
Smoopit
Congrats on an amazing launch! Six months of dedication really show in the product quality. 10,000 developers in beta is impressive. @gateone