Marketplace

Status

The marketplace is not available in the releases documented on this site (Kaba Console 0.146, kabactl 0.87). There is no listing, payment or leasing in the product today. This page describes the idea and the parts of the product that already point toward it.

Every adapter you train is an expert. Experts are assets. The marketplace is where the LoRAs you build, the datasets you curate and the tools you design can be listed, priced, sold or leased.

Planned	Description
Experts for sale or lease	List an adapter with its manifest and evaluation trail. Buyers verify provenance before using it.
Datasets	Curated training sets, released only through the policy gate, so what you share is exactly what you chose to share.
Tools	Tools you design for your experts, sandboxed and ready to drop into a tool loop.
Breeding	Combine experts into a new adapter that inherits from both.

The thinking behind it, including Matter plugins (algorithms that Read, Write or Perform on your data, shipped as model weights), is in Why Kaba?.

What exists today

These are in the product now and are what a marketplace would build on.

Building block	Where
Adapters are portable files with a manifest. Training writes <name>.mpk and <name>.json side by side.	Models & training
Adapters can be imported by URL, with the original download kept as provenance.	kabactl lora-import
Adapters are checked before loading. They pair with a base by architecture family, and shapes are verified.	Machine learning
A training profile and voice. When you train, you choose how the adapter refers to you and how it should speak to someone else who uses it.	Machine learning
Policies are portable. A .kabap bundle carries a policy, its tool overlay and flows, with references to the models and adapters it expects.	File formats
Curation tools. Boost, suppress, annotate and exclude decide what a dataset contains.	Hippocampus
Public service publishing. A node can announce a service to a public directory.	Protocols

Sharing an adapter today

Without a marketplace, sharing is manual and between people who trust each other:

1. Find the adapter’s two files in <storage>/kaba-engine/loras/.
2. Send them, or host the adapter file and share its URL.
3. The recipient copies the files into the same directory, or runs kabactl lora-import <URL>.
4. The recipient selects it in a policy.

Caution

An adapter trained on your memories reflects them. Review what a training run included, and use the date, domain and topic filters, before giving an adapter to anyone else.

[todo: document listing, pricing, leasing, provenance verification and breeding when the marketplace ships.]