ZAI: Z-Space Local AI
The Z-Space AI Collective is a collaborative initiative that pools member resources to fund, build, and operate high-performance, locally hosted AI models. It is a shared learning environment where members contribute financially to a common infrastructure and, in return, gain hands-on access to local AI systems and a growing body of shared knowledge about running them.
From there, we are exploring applications that are approachable and useful for any community, such as group-based shared knowledge bases, where data from individuals is pooled together to benefit from the collective experience of a community.
What does this look like?
We have a small cluster of consumer grade PCs configured behind a secure gateway. Members of the project have keys and passwords to be able to access a number of different models we have configured.
Members can use their API keys to remotely access the AI models, which is compatible with both user facing apps (e.g. an AI chat on your phone) as well as developer use cases (e.g. coding assistants and agents).
Everything runs locally at Z-Space. Your data doesn't leave Canada or even Vancouver.
Technical Details
We are using the open source LiteLLM project as an AI gateway, running on a Proxmox node where we can host other apps as needed.
We configure users, API keys, and quotas, as well as connecting the models we are tuning for different use cases.
Hardware
We have a heterogenous selection of PCs with some new, used, and donated parts.
- ASUS GX10: 128GB of unified RAM (in kind / on loan hardware)
- Orca: 3090 with 24GB VRAM and 64GB of RAM
- Salmon: 2x 1080i with 11GB VRAM and 32GB RAM
In addition, we have an Open WebUI install, and are experimenting with OIKB for syncing data sources for community knowledge base uses.
Community AI in Practice
The Collective is organized around shared discovery. The expectation is that members will experiment together, document what they learn, and make their results openly available to one another. One of the central purposes of this initiative is to advance every member's technical understanding of local AI infrastructure faster than any of them could alone.
Getting involved
We are currently in an early prototype phase, with a select group of early access. If you are an individual or aligned organization in Vancouver, please get in touch by email zai @ z-space.ca.
If you run a co-working or similar community space anywhere in Canada, let us know if you're interested in a similar setup. Subscribe below for updates on what we are learning.
