About Arble

Computers should understand people.

Software is fragmented across dozens of disconnected tools. We built Arble as an agent runtime that bridges intent and execution—on hardware you already own.

Explore Documentation Read Manifesto
Minimal workspace with Arble running on-device
Our Story

Modern computing forces constant context switching between single-purpose applications. The problem is not model capability—it is the absence of a unified agent runtime.

Arble runs the full agent loop on your local hardware. It maintains session memory, manages permission boundaries, and executes tools directly without third-party servers holding your data.

579
Typed tools
61
Built-in toolsets
One
Unified runtime
The Problem

Context does not transfer.
Work does.

Knowledge workers navigate dozens of isolated apps daily. Every boundary drop resets context and manual overhead.

Minimal photography illustrating fragmented digital workflows

Today's software

  • Fragmented apps & lost context
  • Repetitive manual coordination
  • Opaque background data routing

How Arble solves it

  • Intent-driven agent runtime
  • Persistent local memory across sessions
  • Explicit, user-governed permission gates
Our Philosophy

"AI should amplify what people can do, not subtract what people decide."

— Core principle behind Arble permissions

Software should dissolve into the background. You define intent; the runtime handles tool execution, state tracking, and formatting—leaving final decision-making entirely with you.

Principles

What shapes every decision we make.

Human-first

Automated actions operate strictly within explicit boundaries you set.

Local-first

Keys, context, and the agent loop execute on hardware you own.

Privacy-first

Zero telemetry data sales. Enclave protection for all local credentials.

Transparent AI

Every tool invocation, source link, and write action is fully auditable.

Developer friendly

Extensible via Tool SDK, MCP servers, CLI, and self-hosting infrastructure.

Built for longevity

Architected for steady maintenance and reliable long-term developer APIs.

Product

An agent runtime, not just an app.

Arble connects intent to action across desktop, mobile, CLI, and custom servers.

Arble desktop and mobile ecosystem illustration

Desktop Agent

macOS, Windows & Linux native runtime.

Mobile Companion

On-device agent execution on iOS & Android.

Persistent Memory

Cross-session context that stays local.

Permission Gates

Granular control over write and execute tools.

MCP Directory

Native Model Context Protocol server support.

Tool SDK & CLI

Custom toolsets and command-line execution.

The Team

Built by a focused team.

Portrait of Sriram M.

Sriram M

AI Engineering

Developer with a systems infrastructure background. Focused on building reliable local-first platforms and reducing human–machine friction through intent-first interfaces. Architected Arble's core platform, including the MCP framework, local agent execution loop, permission engine, and tool runtime.

Portrait of Sudharsan S.

Sudharsan S

AI Engineering

AI engineer focused on building intelligent, production-ready experiences that connect users with powerful automation. Designed and built Arble's AI systems, integrations and connector ecosystem, along with the desktop experience and web platform — turning complex automation into everyday workflows.

Questions

Frequently asked questions.

Because context does not travel. Work is scattered across apps that cannot see each other, and the assistant that could join them up does not run where the data lives.

No. Arble runs inside your device's OS sandbox—it does not replace it. It is an agent runtime: it manages tool execution, session memory, and security permissions for AI that acts on your behalf.

Unlike web chat boxes, Arble executes real local tools, persists memory across sessions, and keeps API credentials in hardware enclaves.

Local execution ensures zero network dependency for core logic, instant response times, and complete ownership of your contextual data.

Yes. You can build typed tools via our Tool SDK, connect custom MCP servers, or automate execution using the CLI and API.

Claude, OpenAI, DeepSeek, Kimi, Perplexity and NVIDIA NIM, alongside Ollama for models running on your own hardware. You choose the provider in Settings, and fallback chains mean one provider going down does not take the assistant with it.

Yes. Arble is bring-your-own-key throughout. Secrets resolve from your OS keychain or a secret manager and are referenced by name, so they are never written into a config file you might commit.

The agent loop, your memory, your files and the permission gate all run on your device. Model inference runs wherever the model does: choose Ollama and nothing leaves your machine; choose a hosted provider and only that turn’s prompt goes to it. You decide which, per session.

Linux 5.15+ and macOS 14+ for the desktop runtime, with Ubuntu 22.04 and 24.04, Debian 12 and RHEL 9 tested directly. Windows runs through WSL2 for development rather than production. The mobile app ships for iOS and Android.

The future of software

The next interface won't be menus.
It will be intent.

Bring your own models. Keep your own data. Setup takes four minutes — once it lands.

Get notified Documentation