Private AI · Local-first

Your AI.
Your machine.
Your rules.

Local Agent X is a full agentic operating layer that runs on your own computer. No subscriptions. No cloud lock-in. No terminal required. Just an agent that actually does the work.

Download now → See how it works
One-button install
Zero terminal
Your data, your disk
Open source — star it on GitHub★
Local Agent X
97%memory recall · LongMemEval
lax: build the landing page
→ assets found
→ layout generated
→ ready for review
Why local-first

Built for operators, not tinkerers.

Cloud agents are leaky, expensive, and disappear when the API key dies. Local Agent X flips the model — the brain lives on your machine, you own the memory, and the bill stops at your electricity.

⚡

Run anywhere

Native installers for Windows and macOS. No Docker. No Python. No "open a terminal and pray." Double-click and go.

🔒

Truly private

Your files, chats, memory and secrets stay on disk. You choose when (or if) anything talks to a cloud model.

🧠

Persistent memory

Remembers your projects, your people, your preferences — across sessions, across model swaps. Benchmarked, not promised.

97.0%Recall@5 on LongMemEval-S
500benchmark questions, 6 reasoning types
100%knowledge-update recall — 78 of 78 correct
0API keys — embeddings run fully local

Measured on the public LongMemEval-S benchmark with the shipped local embedding model. Reproducible from the repo with one command.

Agent at work
Execution loop

From idea to shipped asset.

Local Agent X is designed around execution loops, not static Q&A.

1

Give the goal

Ask for a build, report, research task, browser action, app edit, or automation.

2

Agent chooses tools

It reads files, opens pages, creates assets, delegates work, or calls APIs as needed.

3

Verify and adjust

Every action gets checked against the outcome instead of assuming success.

4

Deliver the result

You get the file, app, report, answer, or next clear blocker — no vague promises.

Security first
Trust layer

The Ari Kernel watches every tool call.

A built-in security layer reviews every action before it touches your disk, your accounts, or the network. Approve-once, deny-forever, scoped permissions — you stay in the chair.

  • Per-tool allow / deny policy — default-deny on anything new
  • Secrets vault with origin binding — values never reach the model
  • Red-teamed exfiltration containment — blocks 90%+ of a compromised agent's data leaks (98% of credentials) even hidden in base64 / hex / gzip
  • Egress guards on every off-box channel — SSRF + DNS-pinning, outbound secret / PII scan, canary tripwires
  • Untrusted content (web pages, API, MCP & database output) sanitized + flagged before the model sees it
  • Tamper-evident audit log of every action
Capabilities

One agent. Everything you do.

From morning research to closing the books — LAX comes with the tools, the integrations, and the muscle to actually finish work.

🌐

Real browser control

Drives a real browser window — navigate, click, fill forms, read pages.

📁

File & code editing

Reads, writes, and refactors files in your local workspace.

🔐

Secrets vault

Encrypted key storage with per-secret approval — values never reach the model.

🗓

Calendar & email

Reads, drafts, and schedules through built-in calendar and email tools.

🛠

Build apps inline

Creates working web apps directly in your workspace, pinnable to the sidebar.

🎙

Voice in & out

Local Whisper speech-to-text, natural voices back — push-to-talk or hands-free.

📅

Scheduled missions

Recurring agent runs on a cron schedule — nightly reports, daily digests.

🤖

Multi-model

Anthropic, OpenAI, Grok, Gemini, or fully local via Ollama — swap on demand.

In the field

Looks the part. Acts the part.

Operator Agent Shield
Available now

Stop renting your AI.
Own the agent.

Local Agent X is live. Grab the installer for your platform and put an agent on your machine today.

Download for Windows Download for Mac

Windows 10/11 · macOS 13+ · Free to install