Power Performance And Precision

DeadwoodSoftware as a Craft

Plug Deadwood into your product and train your own model on your data. You decide how the model improves over time.

An auditable token layer keeps usage predictable—so you always know what you are spending and what you are earning with a ledger everyone can trust.

chart · crossing.py

Ship models like code, store them like data

Define Your Models

Think of your AI models the same way you think about your application code. You add a small config file called `.deadwood.yml` next to everything else in your repository. In plain language, it lists your models, what data they learn from, and which model should be running in production when someone uses your product.

When you are happy with a change, you commit and push like normal. There is no separate “deployment button” for the model: updating production is part of the same git workflow your team already knows. That keeps releases predictable and easy to roll back if something looks wrong.

The trained model files (the weights) are stored on a decentralized network, so copies live in many places instead of on one company's hard drive. If one host disappears, the data can still be fetched from elsewhere, so you are not betting everything on a single machine staying online forever.

Whenever production switches from one model version to another, that change is written to a permanent record. Later you can see what was live and when—helpful for debugging, compliance, or just explaining to your team what shipped last Tuesday.

  • The ledger keeps a permanent, timestamped record of production changes—who pointed traffic at which model and when—so audits do not rely on chat logs or fuzzy memory.
  • Every model change is version control: branches, tags, and merges work the same way they do for your code, so you always know which definition shipped with which release.
Ocean context · 8-bit chart

Same mental model as sea-scale hosting (think DigitalOcean-style regions): anchors stand for grounding—manifests, commits, and evaluation tied to real artifacts. The treasure on the floor is the upside: publish models to the marketplace and capture payouts when others ship them. Currents still route traffic; .deadwood.yml stays your chart, and the ledger logs every promotion.

Pixel illustration inside a browser window: night sky, animated ocean waves, beach sand, and an anchor partially buried in the sand with gentle motion. The tab reads Deadwood Software; the URL bar shows the standard training path on train.floppydisk.cc. A status strip labels standard features, tensor checkpoints, and commit lineage, and notes that weights pin to commits with inference routing through the promoted checkpoint.

Premium

Deadwood Premium: Standard Features

Models that learn from every interaction.

Deadwood Premium is a cybernetic learning system. It turns your user feedback into self-improving models, automatically.

Cognitive Intelligence

Transparent, Pay-for-What-You-Use

Billing ocean · tokens → treasury
  • → You integrate Deadwood with your API key.
  • → Every prediction, training run, and model swap is counted.
  • → Usage updates live on your dashboard.
  • → Monthly bill = base tier + token overages (if applicable).
  • → Snowflake query fees stay with your Snowflake bill on Premium.

Example month (Premium)

Base: $99/month
├─ 60,000 API calls (within 100K)
├─ 4 trainings (within 10)
└─ Snowflake-connected jobs within quota
Overage charges:
├─ 25,000 extra API calls @ $0.001 = $25
├─ 3 extra trainings @ $50 = $150
└─ Extra prediction tokens @ metered rate = $40
Total: $99 + $25 + $150 + $40 = $314

Transparent. No surprises.

Snowflake Premium

One login for trains, models, and routing—tier-managed billing and access.

Connect repos, issue keys, and ship versions from the same workspace—local experiments through production.