# Scalarware

> One API for language inference, curated image and video generation, deployment
> templates, and live GPU compute. Metered per request against a prepaid balance
> with no markup on published model rates.

This file is a short index. The complete reference, written to be read end to end
by a model, is llms-full.txt.

## Documentation

- [Complete AI-readable documentation](https://docs.scalarware.com/llms-full.txt): every endpoint, field, limit and billing rule
- [Human documentation](https://docs.scalarware.com/)
- [Media skill](https://scalarware.com/skills/scalarware-media/SKILL.md): generate images and video
- [Augustus skill](https://scalarware.com/skills/scalarware-augustus/SKILL.md): publish finished media to a creator account
- [Install both skills](https://scalarware.com/skills/install.sh): one command for Claude Code and Codex

## API

Base URL `https://scalarware.com/api/v1`, bearer `$SCALARWARE_API_KEY` on every
request. Keys carry scopes: `inference:*` for models and media, `compute:*` for
GPUs. Model ids are the market's canonical slugs, for example `z-ai/glm-5.3-flash`.

- Language models: `GET /models` (ordered by real weekly usage)
- Chat inference: `POST /chat/completions` (OpenAI-compatible, supports `stream`)
- Media catalog: `GET /catalog/search?modality=image|video`
- Image generation: `POST /images/generations`
- Video generation: `POST /videos/generations`
- Job lifecycle: `GET|DELETE /media/jobs/{id}`, `GET /media/jobs/{id}/output/{index}`
- GPU offers: `GET /compute/offers` (add `?template={id}` for machines a template runs on)
- Templates: `GET /templates`
- Quote a deployment: `POST /deployments/quote`
- Launch, list, inspect, destroy: `POST|GET /deployments`, `GET|DELETE /deployments/{id}`
- Wait for a launch: `GET /deployments/{id}/wait?since={state}` (long-polls, returns a `tell_user` line)
- Keys: `GET|POST /keys`, `DELETE /keys/{id}`
- Wallet and usage: `GET /wallet`, `GET /usage`

## Behaviour worth knowing

- Requests are priced before they run; an underfunded workspace gets `402 insufficient_balance` with `required_microusd`.
- GPU runtime is charged every minute, and a machine whose workspace runs out of credit is stopped automatically.
- After launching, `status` describes the machine and `setup` describes the model on it. SSH works before the model is ready; a first launch spends 5 to 15 minutes downloading weights.
- Only machines at 1 Gbps or faster and RTX PRO 6000 or better are offered.
- GPU prices returned by the API are the final customer price.

The skill is optional. Any HTTP client can call the API directly.