Docs / Getting started / Quickstart

Quickstart

Install the SDK and issue a tool-using run against a versioned context manifest.

Interface preview. These examples specify the intended developer surface. They are not live endpoints yet.

Prerequisites

Python 3.10+ or Node.js 20+
A Lucci Labs API key when the preview opens

Install

Choose the SDK that matches your project.

pip install lucci

Create your first run

A run combines an instruction with a context manifest. The manifest keeps task information and evidence separate so both can be inspected later.

from lucci import Lucci

client = Lucci()

# Combine an instruction with a versioned context manifest
run = client.runs.create(
    instruction="Summarize the evaluation findings",
    context={
        "task": {"audience": "engineering", "format": "brief"},
        "evidence": ["file_eval_01", "file_eval_02"],
    },
)

print(run.output)

What the SDK is doing

instruction

The immediate objective for this run.

context.task

Output constraints and audience information for the current task.

context.evidence

Versioned files or records made available to the run.

run.output

The final result with tool activity and source references attached.

Use the docs with an AI coding agent

Documentation is available through machine-readable discovery files so a coding agent can inspect the current interface without scraping the rendered page.

https://luccilabs.xyz/llms.txt
https://luccilabs.xyz/docs
https://luccilabs.xyz/openapi.json