Omni

Free desktop app for Windows, macOS & Linux

Build AI agents.
For any task.

Omni is a desktop application that gives your AI agent full computer control — execute commands, read and write files, call APIs, and automate workflows. Connect to 21+ channels, choose from 6 LLM providers, and extend with sandboxed plugins. Everything runs locally on your machine with every action permission-gated.

Desktop app for Windows, macOS & Linux
Build AI agents for any task
21+ messaging channels (Discord, Slack, Telegram & more)
29 built-in native tools
6 LLM providers (OpenAI, Anthropic, Gemini, Ollama & more)
Open source & fully local — your data stays yours

How It Works

Your agent thinks, acts,
and responds — automatically

When a message arrives on any connected channel, Omni's agent loop takes over. It scans the input for safety, reasons with your chosen LLM, calls tools when needed, and sends a response — all in seconds.

01

Message arrives

A user sends a message on Discord, Telegram, Slack, or any of your 21+ connected channels

02

Guardian scans input

The 4-layer security pipeline checks for prompt injection and adversarial attacks before anything reaches the LLM

03

LLM reasons & calls tools

Your chosen AI model (GPT-4, Claude, Gemini, or local Ollama) processes the message and decides which tools to use

04

Tools execute in sandbox

Built-in tools and WASM extensions run in isolated sandboxes — file access, HTTP requests, and more, all permission-gated

05

Response sent back

The final answer is scanned for safety, then delivered to the user on the same channel — in real time via streaming

omni agent

[discord] user: What's the weather in NYC?

 

guardian Input scan CLEAN 2.1ms

llm Routing to gpt-4o via OpenAI

llm Tool call: weather.get_weather({location: "NYC"})

 

perms network.http → api.openweathermap.org ALLOW

wasm Executing in sandbox mem: 12MB / 64MB

wasm Tool result: 72°F, partly cloudy 340ms

 

guardian Output scan CLEAN

llm Streaming response...

 

[discord] omni: It's currently 72°F and partly

cloudy in New York City.

 

Session saved · 847 tokens · 1.2s total

Channels

One agent, 21+ platforms

Connect your AI agent to every platform where your conversations happen. Each channel runs natively — not as a plugin — with full support for messages, groups, and reactions.

Discord
Telegram
Slack
WhatsApp
Microsoft Teams
Matrix
IRC
Google Chat
Signal
iMessage
Line
Twitch
Mattermost
Feishu
Nostr
Nextcloud Talk
Zalo
BlueBubbles
Synology Chat
Twitter / X
Urbit
WebChat

All channels are built into the desktop app — no extra plugins or subscriptions required.

Use Cases

What can you build?

Omni agents handle real work — from answering customer questions to automating entire workflows across every channel.

Customer Support

Build an agent that answers customer questions across Discord, email, and webchat using your docs as context.

Community Moderation

Moderate Discord servers, Telegram groups, and Twitch chat — enforce rules, answer FAQs, and manage conversations automatically.

Email & Message Triage

Let your agent read, categorize, and draft replies to incoming messages across Slack, Teams, email, and more.

E-commerce Automation

Handle order inquiries, track shipments, process returns, and update customers across WhatsApp and webchat.

Knowledge Base Agent

Create an agent that searches your files and documentation to answer questions on any connected channel.

Workflow Automation

Chain tools together — fetch data from APIs, process files, run scripts, and post results to any channel.

Developer Automation

Run tests, debug failures, search code symbols, manage git branches, and spawn sub-agents for parallel development tasks.

AI Providers

Bring your own AI — or run it locally

Omni works with 6 LLM providers out of the box. Use cloud APIs, run models locally with Ollama for full offline privacy, or connect any OpenAI-compatible endpoint. Switch providers any time — your agent keeps working.

OpenAI
GPT-4o, GPT-4, GPT-3.5
Anthropic
Claude Opus, Sonnet, Haiku
Google Gemini
Gemini 2.5 Pro, Flash
Ollama
Llama, Mistral, Phi & more
AWS Bedrock
Claude, Llama, Mistral
Custom
Any OpenAI-compatible endpoint

API keys are stored in your OS keychain (Windows Credential Manager, macOS Keychain, Linux Secret Service) — never in plaintext.

Built-in Tools

Full computer control — sandboxed and permission-gated

29 native tools give your agent full system access — shell commands, file I/O, web requests, git, testing, debugging, code intelligence, interactive REPLs, and desktop app automation. Connect external tools via MCP. Every action is permission-gated, sandboxed, and audited.

Execute Commands
Run shell commands, scripts, and programs directly from your agent
Read & Write Files
Read, create, edit, and patch files — logs, configs, documents, code
Grep & Search
Regex search across files and directories with line numbers and context
HTTP & Web Search
Call any API, fetch web pages, search the web, and scrape content with anti-bot stealth
Persistent Memory
Save and search agent memory that persists across sessions with tags and categories
Git Version Control
10 structured actions — commit, branch, merge, diff, stash — with automatic secret scanning
Test Runner
Auto-detect and run tests — Rust, Jest, Vitest, pytest, Go, .NET — with structured pass/fail results
Code Intelligence
Offline symbol search across 9 languages plus real-time LSP for go-to-definition, references, hover, and diagnostics
Debugger (DAP)
Launch, attach, set breakpoints, step through code, and inspect variables via Debug Adapter Protocol
Interactive REPL
Persistent Python and Node.js sessions with state between executions — up to 3 concurrent
Sub-Agents
Spawn parallel sub-agents with their own context and tools for concurrent task execution
MCP Client
Connect external MCP tool servers via stdio — tools are auto-discovered, namespaced, and Guardian-scanned
App Automation
Launch and control desktop apps — click, type, read, screenshot — with LOLBIN blocklist and password protection
Clipboard
Read from and write to the system clipboard for quick data transfer
Notifications & Cron
Desktop notifications for alerts and cron scheduling for recurring automated tasks

Every tool action requires explicit user permission. Agents can't access files, run commands, or make network requests without your approval.

Security

4-layer security pipeline

Every extension passes through four independent analysis layers before it can be published. No exceptions.

01

Signature Scanning

30%

79+ malicious patterns matched against compiled binary

02

Heuristic Analysis

25%

Behavioral rules flag suspicious permission combinations

03

AI Code Review

30%

Claude analyzes code semantics for hidden threats

04

Sandbox Testing

15%

Isolated execution detects runtime anomalies

omni scan

$ omni scan weather-tool.wasm

 

Running security pipeline...

 

PASS | Signature scan0 matches

PASS | Heuristic analysisscore 95/100

PASS | AI code reviewscore 92/100

PASS | Sandbox testingscore 98/100

 

Verdict: CLEANoverall 96.3

Extension approved for marketplace.

Why Omni

All the power.
None of the risk.

Other AI agent tools give you raw computer control with no guardrails. Omni gives you the same power — shell access, file I/O, API calls, process execution — but every action is sandboxed, permission-gated, and audited. Full control for your agent, full visibility for you.

Persistent Desktop Agents

Agents run continuously on your machine, listening across 21+ channels and responding in real time — not one-shot prompts, but always-on automation.

Full System Access

Execute commands, read/write files, make HTTP requests, manage git repos, run tests, debug code, and control desktop apps. Your agent can do anything you can — with your permission.

Deny-by-Default Security

Every capability is blocked until you explicitly allow it. File access is path-scoped. Network requests are domain-scoped. Commands are whitelisted. Full audit trail.

Guardian Anti-Injection

A 4-layer pipeline (signatures, heuristics, ML classifier, policy validation) scans every input and output for prompt injection attacks in real time.

Ready to build your AI agent?

Download Omni for free on Windows, macOS, or Linux. Set up your first AI agent in minutes — no coding required.