Quickstart

Start from the surface that fits your workflow.

CLI for local runs, SDK for Python pipelines, MCP for agents, and HTTP for queued jobs.

Run paths

Four entry points, one JSON contract.

CLI

paralleliz --input examples/story_auto.json --review \
  --validation-report out/review/story_auto_report.json

paralleliz --input examples/story_auto.json \
  --output out/decks/story_auto.pptx

Python SDK

from visuals_core.sdk import validate, render

result = validate(deck)
render(deck, "out/decks/story_auto.pptx")

MCP

paralleliz mcp serve --stdio

HTTP job service

uv run python visuals_service.py --host 127.0.0.1 \
  --port 8765 --jobs-dir jobs

Submit specs through `POST /jobs`, poll the job state, then download artifacts from the service instead of reading local paths directly.