Quickstart

Boot plnt locally in under 60 seconds. No cluster, no GPU, no signup.

1. Install

git clone https://github.com/plnt-work/plnt && cd plnt
python3.12 -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"

The .[dev] extra pulls pytest and ruff. Skip it if you’re just running.

2. Start the playground

plnt playground up

You’ll see:

plnt playground on http://127.0.0.1:8080 — mock backend, ctrl+C to stop.
  workflows:  http://127.0.0.1:8080/v1/workflows
  invoke:     POST http://127.0.0.1:8080/v1/workflows/<id>/invoke
  docs:       http://127.0.0.1:8080/docs

The default backend is a mock that runs each workflow step against a stub LLM — enough to verify the wire is intact. Real GPU backends come later (Local dev).

3. Invoke a workflow

In a second terminal:

# list registered workflows
plnt list

# one-shot invocation, streams step events to stdout
plnt run review-responder --input '{"review":{"rating":2,"text":"Waited 40 min..."}}'

# tail the Temporal orchestration for a running workflow
plnt logs review-responder --follow

# copy-paste curl examples
plnt playground curl

Or directly with curl:

curl -s http://127.0.0.1:8080/v1/workflows | jq

curl -s http://127.0.0.1:8080/v1/workflows/review-responder/invoke \
  -H 'content-type: application/json' \
  -d '{"input":{"review":{"rating":2,"text":"Waited 40 min..."}}}' \
  | jq

4. Add your own workflow spec

Point plnt at a workflow spec pulled from any registry (or a local directory):

cat > /tmp/workflows.json <<'EOF'
[
  {"id": "review-responder",
   "ref": "review-responder@1.2.0",
   "registry": "s3://microagents",
   "backend": "mock"},
  {"id": "review-responder-live",
   "ref": "review-responder@1.2.0",
   "registry": "s3://microagents",
   "backend": "http",
   "upstream_url": "http://127.0.0.1:8000"}
]
EOF

PLNT_PLAYGROUND_CONFIG=/tmp/workflows.json plnt playground up

Anything with "backend": "http" proxies to the URL — any workflow runner that speaks the plnt /invoke contract works.

Next

• Local dev — wire the site’s UI to your local API.
• Kind demo — the full K8s + Temporal flow on your laptop.
• Architecture — the 4-layer stack view.