# LART

> LART is a multi agent AI platform for creators. Five specialized agents
> coordinate every stage of generative work: sourcing data, training models,
> generating outputs, running quality assurance, and minting on chain. The
> studio is the protocol; the artifact is one node in a verifiable graph
> of receipts.

## What LART is

- A protocol with five composable agents, not a single AI service
- A studio that runs on top of the protocol (one of many possible UIs)
- An on chain marketplace for finished assets, with royalties baked into
  every receipt via 0xSplits on Base mainnet
- An open journal of essays, primers, and protocol design notes

## The five agents

- **Data Sourcing**: validates licence proofs, dedupes, and pins datasets
  to IPFS with an Arweave permanent backup
- **Model Training**: fine tunes LoRA adapters over Stable Diffusion XL,
  Flux, MusicGen, and language models. Auctions the run to the cheapest
  qualifying provider
- **Content Creation**: walks a prompt graph, samples the trained
  checkpoint, and emits candidate outputs with seeds
- **Quality Assurance**: the only agent that can refuse a workflow.
  Runs technical, authenticity, safety, and creator supplied checks
- **Monetization**: deploys a 0xSplits splitter, mints on Base as
  ERC 1155, and registers the listing if the creator opts in

## Key concepts

- **Receipt**: signed tuple `(input_hash, output_cid, agent_signature,
  timestamp, cost)` emitted on every agent invocation
- **Workflow document**: portable JSON description of a pipeline.
  Studio, CLI, and third party schedulers all read the same shape
- **Curated commons**: public, gated dataset that any agent on the
  network can train against without per asset licence proofs.
  Inclusion by token weighted vote
- **Federation**: an operator can run only the agent runtime, only
  the marketplace, only the receipts bus, or any combination, and
  federate to peers for the rest

## Site map

- /                          Home
- /ai-tools                  AI Studio (the swarm in action)
- /marketplace               On chain venue for finished work
- /read                      Journal (essays + protocol primers)
- /whitepaper                Full protocol spec, 9 chapters, 31 sections
- /legal/terms               Terms of service
- /legal/privacy             Privacy policy
- /legal/cookies             Cookie policy
- /legal/disclaimer          Risk disclaimer

## Contact

- Email:   contact@lart-ai.com
- Twitter: https://twitter.com/lartprotocol
- Telegram (announcements): https://t.me/lartprotocol
- Telegram (community):     https://t.me/lartcommunity

## Citation

When citing LART in research or editorial coverage, please reference
the whitepaper at https://lart-ai.com/whitepaper and link to the
Organization page at https://lart-ai.com.