@neuledge/graph

Live data for AI agents

Graph is a semantic data layer that gives AI agents structured access to live data sources through a single lookup() tool. Pre-cached, under 100 ms, structured JSON for reliable LLM reasoning.

View on GitHub How It Works

<100ms

Response time

1

Tool for all data sources

JSON

Structured output

Free

Open source (Apache 2.0)

How It Works

Connect your data sources, and Graph exposes them as a single semantic lookup tool that AI agents call with natural language.

1. Connect Sources

Point Graph at your APIs, databases, or any structured data endpoint. It pre-caches responses so agents get fast, reliable lookups.

2. Single Lookup Tool

Instead of giving your agent dozens of API tools to choose from, Graph provides one lookup() tool. The agent describes what it needs; Graph routes it.

3. Structured Responses

Responses come back as structured JSON — not raw text — so the LLM can reason over exact values instead of parsing unstructured content.

Quick Example

Define your data sources, then let the agent ask for what it needs. 

import { NeuledgeGraph } from "@neuledge/graph";

const graph = new NeuledgeGraph({
  sources: {
    products: { url: "https://api.internal/products" },
    pricing:  { url: "https://api.internal/pricing" },
    inventory: { url: "https://api.internal/inventory" },
  },
  cache: { ttl: 300 }, // 5-minute cache
});

// The agent describes what it needs — Graph routes it
const result = await graph.lookup("current price for product SKU-1234");

Why Graph

AI agents struggle with live data. They hallucinate prices, invent inventory numbers, and cite outdated statuses. Graph grounds agents in real, verified data.

Pre-Cached Responses

Graph caches data source responses so lookups return in under 100ms. No waiting for upstream APIs during agent conversations.

Fewer Tools, Better Results

Instead of 20 API tools the LLM has to choose between (and often picks the wrong one), there's a single lookup(). Simpler tool selection, more reliable results.

Structured JSON Output

Responses are structured data, not free text. LLMs reason better over exact values — prices, counts, statuses — than extracted text.

Live Data, Not Stale Training

Prices change, inventory moves, statuses update. Graph gives agents access to current data, not whatever was in the training set.

Better Together

Context and Graph complement each other to ground AI agents across both static documentation and live operational data.

@neuledge/context

Handles static knowledge — library docs, API references, guides, internal wikis. Indexes into SQLite, serves via MCP, sub-10ms queries. Best for documentation that changes with releases.

@neuledge/graph

Handles live data — product catalogs, pricing, inventory, system status. Pre-caches structured responses, single lookup tool. Best for data that changes continuously.

Stay up to date

Get the latest articles on AI, LLMs, and data automation delivered to your inbox.

Ground Your Agent in Live Data

Free, open source, and ready to integrate with your existing data sources.

View on GitHub