ENA — open AI, trained by everyone

ENA turns Animica mining into useful AI work. One command runs a miner that earns ANM for proof-of-work and for AI: contributing data, training a shared model, serving it for inference, and (on qualified GPUs) Bittensor. Everything settles to your Animica address.

Training pools →Download a GUI miner
pip install --upgrade animica
animica up

Train together (pools)

Many contributors fund and train ONE model in rounds. The dataset is sharded across trainers; verified work is rewarded proportionally.

Serve while training

The promoted checkpoint is served over an OpenAI-compatible API while the next round trains. Servers earn per token, in ANM.

One global model

Every pool — run by anyone — converges on a single canonical model: a shared on-chain head all servers serve and all clients use.

Useful-work mining

CPU miners also do scrape / clean / embed / eval jobs; each job is receipted and paid. Mining funds the model.

Coding agent

Use the pool model as a sandboxed coding agent (`animica ena code`) over any OpenAI-compatible endpoint.

ANM-only rewards

PoW, useful-work, training, serving, and Bittensor all pay out in ANM to your address. No external payout to manage.

How it fits together

  1. Run animica up — it auto-creates a wallet and mines + does AI by capability.
  2. Fund a training pool to accelerate the global model.
  3. GPU miners train shards and serve the promoted checkpoint; qualified GPUs also serve Bittensor.
  4. Build with the model via the OpenAI-compatible endpoint or the ENA coding agent.