AI operating layer
Infrastructure for
intelligent systems.
Gateway, memory, retrieval, agents and inference. Five products that fit together, from a lean team that ships.
<10ms
speed-first performance
1
simple SDK for everything
5
products, one stack
100%
owned, self-hostable
Index
Five products. One operating layer.
Each product stands alone. Together they form one stack.
- 01
NAIOS
AI gateway & agent OS control plane
One gateway for every model, and a control plane for every agent.
- 02
NAIRAG
Ultra-fast graph RAG engine
Graph retrieval with anomaly detection, from ingest to search.
- 03
NAIMEM
Real-time context engine for agents
Remember, recall and forget. Working memory your agents reach in milliseconds.
- 04
NAIGENT
AI agent app, multi-agent, full automation
Agents that do the work, for individuals and small businesses.
- 05
NAIDOT
LLM, SLM & inference infrastructure
Inference setup service and the lab behind NAI's own models.
How it fits
One stack, no glue code.
NAIOS routes. NAIRAG retrieves. NAIMEM remembers. NAIGENT acts. NAIDOT runs the models.
Fast by default
Latency is a feature. Every component is measured in microseconds before milliseconds.
Simple APIs
One call to plug in. No SDK sprawl, no ceremony.
Own the stack
From gateway to model weights. Nothing you can't inspect, host or replace.
Developer surface
Three calls. That's the memory API.
NAIMEM is what an agent's working memory should feel like: instant, scoped, and boring to integrate.
import { Memory } from "@nai/mem"
const mem = new Memory({ scope: "user:42" })
await mem.remember("prefers concise answers")
const ctx = await mem.recall("user preferences")
// → ["prefers concise answers"] 2.1 ms
await mem.forget({ olderThan: "7d" })
NAI Lab
We build the layer others rent.
Gateway, retrieval, memory, agents and the models underneath. Owned end to end by a lean engineering team.