Quickstart — Get Started

This quickstart guides you through initializing the NeuronScope AI Engineering Platform, attaching an observability scope to an application, running standardized evaluations, profiling execution latency, and inspecting sealed telemetry artifacts.

Step 1: Install Platform

terminal

bash

$pip install "neuronscope[all]==0.9.2"
$ns doctor
python 3.12 · neuronscope 0.9.2 · AI Kernel ok ·adapters: [torch, jax, llm_api, vector_store]

Step 2: Initialize & Observe

NeuronScope attaches to models, RAG pipelines, or autonomous agents using non-invasive context managers.

quickstart_ai.py

python

import neuronscope as ns

# Initialize AI Engineering Platform instance
platform = ns.Platform.init(app_name="quickstart_service")

# Observe execution, latency, and tokens across system components
with ns.Scope(platform, monitor_all=True) as scope:
    response = platform.run_agent("support_agent", input="Analyze latency metrics")

# Retrieve telemetry & save sealed artifact
telemetry = scope.get_telemetry()
print(f"Latency: {telemetry.latency_ms}ms | Cost: ${telemetry.cost_usd}")
scope.save("quickstart_run.nsz")

Step 3: Evaluate System Quality

Run standardized evaluation pipelines to measure accuracy, hallucination scores, and latency.

eval_quickstart.py

python

from neuronscope import Evaluator, load_artifact

artifact = load_artifact("quickstart_run.nsz")
result = Evaluator.evaluate_hallucinations(artifact)

print(f"Hallucination Score: {result.score} (Status: {result.status})")

Step 4: Inspect Telemetry & CLI

terminal

bash

$ns inspect quickstart_run.nsz
app         quickstart_service
fp:model    fp:model:b3c2f0a1e8d7a2c6...
fp:prompt   fp:prompt:8f10a7b2c4e6...
telemetry   12 execution steps · 4 tool calls · 1.2 MB

Where to Go Next

Explore Fingerprinting to understand content-addressed identity.
Learn about System Architecture and the 8-layer platform model.
Read the Compare Two Models recipe for production QA.