Skip to main content
This walkthrough takes you from zero to a live decision feed showing real LLM tool calls flowing through Inviolet’s intent extraction — without any policy or blocking enabled.
This is Phase 1 — Anonymous Observation. No identity, no policy, no blocking. The gateway watches and learns. You can flip on enforcement later.

​
Prerequisites

  • An Inviolet workspace (sign up free)
  • Node 20+ or Python 3.10+ for SDK install
  • Five minutes

​
1. Install the SDK

npm install @inviolet/sdk

​
2. Wrap your LLM client

import { InvioletGateway } from "@inviolet/sdk"
import Anthropic from "@anthropic-ai/sdk"

const inviolet = new InvioletGateway({
  apiKey: process.env.INVIOLET_API_KEY!,
})

const anthropic = inviolet.wrap(
  new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY! }),
)

// Use anthropic exactly as you would normally — every tool call now
// flows through Inviolet's intent extractor.

​
3. Make a tool call

Run your application normally. Every LLM tool call your agents make will appear in the decision feed at app.inviolet.ai/decision-feed.

​
4. Watch the heatmap fill in

After 50–100 tool calls, the Query Pattern Map will start to show what your agents actually do — which tables they read, what columns they touch, what intentions they declare. This is the foundation for declaring policy in Phase 3 (“Policy + LLM lockdown”).

​
What’s next

If you get stuck, the Inviolet AI assistant is on the marketing site — it can search this documentation and book a session with the team.