Quickstart
Quickstart
Deploy one reusable Base and connect a new AI product to it.
This section only does one thing: help you run a complete VisibleBase call chain as quickly as possible.
What you end up with is not just a demo page. It is the smallest product-shaped system:
- one reusable Base
- one model record
- one
product_id - one
user_tokenthat an end user can actually use - one
UserClientthat successfully calls Base
You do not need a separate AI backend for every new product. A new project can start as a client first and connect to an already deployed Base.
What you will end up with
Product frontend / extension / appYour product only needs
product_id + user_token so you can ship UX first.UserClientReads
models() and then calls text(), stream(), image(), or another service.VisibleBase BaseVerifies the token, reads the model directory, then chooses the real execution path with
default() + match().Real providerOpenAI, Claude, Gemini, private models, or your internal AI Gateway.
Recommended path
- Use it now: run the shortest end-to-end path with
products/serverandproducts/client - Create Base: create Base and let the first startup initialize the default database automatically
- Add model: write model records as database data
- Issue
user_token: give a product user a valid call credential - Connect client: call Base from the product with
UserClient
Which path should you choose
I only want one working flow first
Starting from the templates is the fastest option. Verify Base, product, token, and UserClient first, then come back for the details.
I want to start in my own projectStart directly from new Base() and wire the database, models, tokens, and client into your own repository.
Assumptions in this quickstart
- the default database path is
./.base/visiblebase.sqlite - models come from the database, not runtime registration code
- all products currently share the same set of active models
- service-level query fallback is defined in code with
default()