Documentation

Alys runtime architecture.

Alys is a local-first autonomous dataset generation engine. The cloud coordinates accounts and metadata; the intelligence runs on the user's machine.

CLI Runtime

Generate locally.

npm install -g alys-akusa

alys generate "B2B SaaS sales objections"

pnpm alys -- generate "B2B SaaS sales objections"

pnpm alys -- generate "AI evaluation datasets" --format jsonl,csv,rag

Use --rows 1000 for larger test datasets, or raise --sources when you want broader source coverage.

Launch access

5 free generations.

Every account starts with five free dataset generations while payments are offline.

Accounting

1 dataset = 1 generation.

A run requesting five datasets spends five generations and uses the full free allowance.

NPM

alys-akusa

The CLI publishes under the Akusa scope because the unscoped package name is already taken.

apps/alys-web

Dashboard, auth, telemetry, dataset job metadata.

apps/cli

Local CLI entrypoint for `alys generate` and local dev via `pnpm alys`.

packages/agents

Discovery, extraction, verification, debate, dedupe, structuring, curation.

packages/orchestration

Swarm coordinator and event stream.

packages/storage

Local-first filesystem workspace under ~/Alys.

packages/datasets

JSONL, CSV, markdown, instruction, and RAG serializers.

Swarm Agents

Research is a pipeline, not a scrape.

DiscoveryAgent ranks sources by relevance and source quality.

ExtractionAgent cleans HTML and converts content into text/markdown.

VerificationAgent compares findings and scores confidence.

DebateAgent challenges contradictions before curation.

DeduplicationAgent removes semantic overlap.

StructuringAgent converts findings into dataset records.

DatasetCuratorAgent exports final artifacts and metadata.

Local Workspace

~/Alys/
  datasets/
  crawls/
  embeddings/
  exports/
  cache/
  logs/

Future cloud sync can be added behind this abstraction, but the MVP keeps generation artifacts local by default.

Create dataset job