The Operating System for AI Agents

LLMBasedOS

The infrastructure layer that turns any LLM into an autonomous agent with persistent memory, event-driven workflows, and real-world actions.

Like Linux for AI — giving Claude hands, memory, and the ability to act over time.

Get StartedWhy Claude Code?Documentation

The Problem

LLMs are stateless chat interfaces

Today's AI can reason brilliantly, but it can't act persistently. Every session starts from zero. No memory. No agency. No autonomy.

It's like having a genius friend with amnesia who can only talk but never do.

The Solution

LLMBasedOS gives LLMs superpowers

🧠

Persistent Memory

Vector store that survives sessions — remembers context forever

⚡

Event-Driven Workflows

PubSub system that triggers actions on real-world events

🌐

Browser Control

Full Playwright automation — click, type, screenshot, scrape

🔌

40+ MCP Methods

KV store, SQLite, mail, alerts, trading, and more

Architecture

A middleware layer for AI agents

LLMBasedOS sits between your LLM and the real world, providing the primitives agents need.

┌─────────────────────────────────────────────────────────┐
│  Claude Code / GPT / Gemini / Any LLM                   │
└────────────────────────┬────────────────────────────────┘
                         │ MCP Protocol (JSON-RPC)
                         ▼
┌─────────────────────────────────────────────────────────┐
│  Gateway (auth, routing, rate-limits)                   │
└────────────────────────┬────────────────────────────────┘
                         │ Unix sockets
                         ▼
┌─────────────────────────────────────────────────────────┐
│  Arcs (microservices)                                   │
│  ┌────────┐ ┌────────┐ ┌────────┐ ┌────────┐ ┌────────┐│
│  │ memory │ │ pubsub │ │browser │ │   kv   │ │  mail  ││
│  └────────┘ └────────┘ └────────┘ └────────┘ └────────┘│
└─────────────────────────────────────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────────────┐
│  Infrastructure                                         │
│  Redis │ ChromaDB │ Playwright │ Ollama │ SQLite       │
└─────────────────────────────────────────────────────────┘

For Claude Code

Turn Claude into a real teammate

Claude Code is powerful. With LLMBasedOS, it becomes operational.

Without LLMBasedOS

✗

"What did we do last week?"

"I don't know"

✗

Waits for you to ask

Passive

✗

Can only touch local files

Limited

✗

Forgets everything on restart

Stateless

With LLMBasedOS

✓

"What did we do last week?"

"You fixed 3 bugs and deployed v2.1"

✓

Watches for events 24/7

"Hey, CI is failing!"

✓

Browse, email, API calls

Full internet access

✓

Remembers context forever

Persistent state

It's not a plugin. It's the difference between a calculator and a co-founder.

Native Integration

Works with Claude Code & Codex

All arcs are exposed as MCP tools — no glue code, no adapters, just native capabilities.

◇

Claude Code

MCP Integration

$ claude mcp add llmbasedos
C

OpenAI Codex

Install as a Skill

$ codex install llmbasedos

All 40+ methods become native tools in your AI assistant

Available MCP Methods
memory_store
memory_recall
pubsub_subscribe
pubsub_emit
browser_goto
browser_screenshot
browser_click
browser_type
kv_set
kv_get
sentinel_run
llm_route
mail_send
sqlite_query
events_publish
fs_list

+ 24 more methods

Core Arcs

The three pillars of agent autonomy

Memory, Events, and Actions — everything an AI needs to operate independently.

🧠

Memory Arc

ChromaDB + Ollama

Persistent vector memory with semantic search. Store context, recall by meaning, build knowledge over time.

memory_storememory_recallmemory_getmemory_delete
⚡

PubSub Arc

Redis-backed

Event-driven workflows with subscriptions. React to signals, trigger callbacks, coordinate agents.

pubsub_subscribepubsub_emitpubsub_historypubsub_list
🌐

Browser Arc

Playwright

Full browser control. Navigate, click, type, screenshot, extract data, run JavaScript.

browser_gotobrowser_clickbrowser_typebrowser_screenshot

KV Store

Redis-backed key-value

SQLite

Local SQL database

Mail

Send emails & alerts

Sentinel

Background watchers

LLM Router

Multi-model routing

Events Bus

Real-time streaming

File System

File operations

+ Custom

Build your own arcs

Get Started

Up and running in 2 minutes

Full Stack

Docker + all services

$ git clone https://github.com/iluxu/llmbasedos.git

$ cd llmbasedos

$ docker compose up -d

View on GitHub

MCP Only

For Claude Code / Codex

# Claude Code

$ claude mcp add llmbasedos

# Codex

$ codex install llmbasedos

Read the docs

Use Cases

What you can build

Real workflows, not demos. Production-ready from day one.

AI Developer Teammate

Claude that remembers your codebase, watches CI/CD, and proactively fixes issues.

  • →Persistent project context
  • →CI failure alerts
  • →Auto-PR creation

Research Agent

Browse the web, extract data, store findings, and build knowledge bases autonomously.

  • →Web scraping
  • →Document memory
  • →Scheduled research

Ops Automation

Monitor systems, react to alerts, execute runbooks, and coordinate responses.

  • →Event subscriptions
  • →Multi-agent coordination
  • →Audit trails

Token

LLMBasedOS Token on Base

The LLMBasedOS token anchors alignment and long-term access primitives.

View on Basescan

Token Contract

0xC1f85EBfC85d09EBEDbd55b2865092944F362cB5

Owner Address

0x6296585f5D80564D0c7e241306d9c19D8cc85F9D

FAQ

Frequently Asked Questions

Ready to give your AI superpowers?

Start building autonomous agents that remember, react, and act.

Get Started on GitHubContact Us