Puras turns a prompt into a typed, testable, deployable skill — and runs it on your own machine.

Why Puras?

You can already fire a prompt at an LLM SDK — or an agentic one that wires up tool calls for you — and get an answer back. The hard part is everything around it: persisting memory across runs, orchestrating long-running, multi-step pipelines, and surviving the expensive failure — the run that burns through a dozen costly steps and then dies on the last one, throwing away everything before it. Add schema-validating the output and testing the whole thing so it doesn't silently regress, and you're maintaining a pile of glue code and infrastructure per prompt. A skill is that prompt promoted to a real unit:

• a typed input/output contract — schema-validated in, schema-valid JSON out, every time;
• memory that persists across runs — a local SQLite store on your laptop, semantic recall on the platform;
• evals that test the prompt like code, with a CI gate;
• one loop, two environments — build against a local API and ship the identical bundle with puras deploy; in production, long runs are checkpointed, so a step that fails near the end resumes instead of re-running everything before it.

If all you need is a single completion, reach for the SDK — Puras earns its place the moment a prompt becomes something you run repeatedly, test, and ship.

Getting started

Install the CLI and the offline runner:

pip install "puras[local]"

A skill is a folder — a prompt and an input/output contract. Create two files. skill.yaml declares the contract as plain JSON Schema — the inputs it takes and the fields it must return:

# triage/skill.yaml
title: Ticket Triage
description: Sort an inbound support message into a category, priority, and summary.
entrypoint: SKILL.md          # markdown entrypoint = agentic; the file is the system prompt

input_schema:
  type: object
  required: [message]
  properties:
    message: { type: string }

output_schema:
  type: object
  properties:
    category: { type: string, enum: [bug, billing, feature_request, other] }
    priority: { type: string, enum: [low, medium, high, urgent] }
    summary:  { type: text }   # `text` = multi-line string

SKILL.md is the system prompt the agent runs with:

<!-- triage/SKILL.md -->
You triage inbound support messages. Read the `message` and respond with its
`category`, `priority`, and a one-line `summary`. Judge priority by user impact,
not by how loud the message is.

Serve it. puras serve exposes the same job API your app will hit in production, backed by the local runner on your own LLM key (if no key is set, the CLI asks for one):

puras serve          # serves ./ → http://127.0.0.1:8787

Now call the skill from your app — point any Puras SDK at the local base URL:

import puras

client = puras.Client(api_key="local", api_base="http://127.0.0.1:8787", skillpack="local")
out = client.run("triage", {"message": "I was charged twice this month and want a refund!"})
print(out["category"], out["priority"])   # → billing high

import { Puras } from "puras";

const puras = new Puras({ apiKey: "local", apiBase: "http://127.0.0.1:8787", skillpack: "local" });
const { category, priority } = await puras.run("triage", { message: "I was charged twice this month and want a refund!" });
console.log(category, priority);   // → billing high

That's the whole loop. When you ship, change the base URL to https://api.puras.co, run puras deploy, and the same app code runs against the managed platform — nothing else changes.

Want to iterate faster? puras run --local triage -i message="…" runs a skill straight from the CLI, puras eval --local gates it on evals, and the examples/ folder has ready-to-run skillpacks.

Open-source vs Cloud

Puras is open-core: the runner — the agent loop and the local tool surface — is MIT-licensed and runs fully offline, forever. The hosted platform at puras.co is how the project is sustainably funded, and it adds the managed surface that can't exist on a single laptop. Premium isn't a crippled core — it's the capabilities that need real infrastructure.

The local runner gives you the free, offline core; the hosted platform adds the managed surface — persistence, scale, durable resume, retrospectives — for when you ship the same bundle unchanged.

Community

Questions, bugs, and skill ideas are welcome in Issues and Discussions. PRs that improve the runner, the docs, or the examples are appreciated.

Working from a checkout instead of PyPI:

pip install -e .             # puras-runner (the runtime)
pip install -e worker/sdk    # the puras CLI + SDK

License

MIT. The hosted platform's server-side code is separate and commercial — this runner, the SDK, and the examples are MIT and yours to use, modify, and self-run.