Skip to main content
Linux & Windows WebGPU

OneStone AI

Your AI.
Your machine.
Your data.

A desktop AI agent platform — chat with live tool calls, scheduled routines, persistent memory, email, and analytics, all running locally. Choose your provider. Keep your data.

v0.7.17

shipping — see releases

105+

built-in agent tools

7

AI providers, your choice

100%

of your data stays local

Everything in one place

Not another chatbot. A complete operations center for AI-augmented work.

Agent Chat

Streaming conversations with inline tool calls, markdown rendering, code highlighting, file attachments, and local-Whisper voice input. The agent doesn't just answer — it reads, writes, searches, and acts, showing every tool call as it happens.

Reach Anywhere — Relay

Pair your phone with one QR scan. Your local dashboard becomes reachable from any network — no port forwarding, no static IP, no domain. Outbound-only WebSocket; your data still lives on your machine.

Scheduled Routines

Cron-driven agent jobs with run history, visual schedule picker, category grouping, and multi-turn continuity.

Persistent Memory

Rules, knowledge, and project context that carry across sessions. After every summarize cycle the agent drafts knowledge proposals you accept or dismiss — memory you curated, not auto-saved noise.

Skill Library

Bundled skills the agent loads on demand — TypeScript, React, Hono, Three.js, security review, document generation, and more. Add your own; they live in ~/.onestone/skills/.

Agent Identity

Define your agent’s identity, operating rules, and personality in a single AGENTS.md file — edited in the dashboard, backed up on every save.

Integrated Email

Thread-based inbox with compose, reply, and forward. The agent can send and receive email on your behalf — routines can email autonomously on schedule.

Analytics & Cost Tracking

Token usage by model, cost breakdowns, tool-call performance, and routine run history. Know exactly what your agent costs.

Memory, analytics, email, and connections — one app

Every capability lives in the same local dashboard. A full walkthrough is on the way.

Memory that persists

Rules, knowledge, and project context carry across every session. The agent learns your preferences, remembers decisions, and builds institutional knowledge over time.

Know what it costs

Uptime, routine counts, memory entries, and project namespaces at a glance. Track token usage by model and conversation. No surprises.

Email built in

Thread-based inbox with compose, reply, and unread tracking. The agent sends and receives email on your behalf — routines can email autonomously on schedule.

Connect your providers

Add API keys for Anthropic, OpenRouter, Groq, Mistral, Gemini, or run local models with Ollama. Visual status indicators show what’s connected at a glance.

Deep MCP tool integration

The agent doesn't just chat — it acts. A full suite of MCP tools gives it real capabilities across your system.

Memory & Knowledge

Save, search, and recall rules, knowledge, and project context. The agent builds institutional memory over time.

Web Search & Fetch

Search the web and read pages in real time. The agent researches, verifies, and stays current without leaving the app.

File Operations

Read, write, browse, and edit files on your system. Bash commands, code edits, and project navigation built in.

Browser Automation

Launch browser sessions, take screenshots, click elements, and fill forms. Visual verification of live sites and apps.

Email Send & Receive

Full email via AgentMail — send, reply, forward, and check inbox. Routines can email on schedule autonomously.

Calendar

Track deadlines, milestones, and time-anchored tasks. The agent knows what's due and surfaces it in conversation context.

Desktop Capture

Screenshot your desktop or record video. The agent can see what's on screen for visual debugging and verification.

Contacts & Projects

Manage contacts and per-project knowledge namespaces. Context loads automatically based on what you're working on.

Built on the Model Context Protocol. Connect external MCP servers to extend the agent.

Configure everything from the dashboard

No config files to hand-edit. Every setting is accessible from the browser-based dashboard.

Provider API Keys

Add, remove, and switch between AI providers with visual status indicators. Connected state shows at a glance.

Model & Reasoning

Select your default model from a dropdown. Override per-routine for cost optimization — a fast model for simple tasks, a frontier model for complex ones.

Scheduler Settings

Concurrency limits, failure thresholds, catch-up windows, retention policies — all live-editable sliders and inputs.

Tool Policies

Enable or disable individual agent tools with toggles. Control exactly what capabilities the agent has access to.

Routine Builder

Create routines with a visual schedule picker — daily, weekly, hourly, or custom cron. Live preview shows exactly when it will fire.

Channel Budgets & Fallback Chains

Daily $ caps per channel with auto-downgrade thresholds, and cross-provider fallback chains so a provider outage rotates to the next in line instead of killing the turn.

Your provider, your choice

Switch between AI providers without changing your workflow. Use the best model for each task.

Run models locally with Ollama, connect to any cloud provider with your own API key, or route Claude turns through your Claude subscription via the official CLI — no per-token billing.

Your API keys stay on your machine. Never sent to us.

Common questions

Straight answers — no marketing haze.

Is OneStone free?

Yes — the desktop app is a free download for Linux and Windows. You bring your own AI provider account (or run fully local models with Ollama), so the only costs are whatever your chosen provider charges you directly.

Where does my data live?

On your machine. Conversations, agent memory, projects, contacts, and analytics are stored locally on your computer. The only network traffic is the API calls to the AI provider you selected — there is no OneStone cloud account and no telemetry harvesting your content.

Which AI providers does it support?

7 providers out of the box: Anthropic, OpenAI, Google Gemini, Groq, Mistral, OpenRouter, Ollama. You can switch the model per channel — chat, routines, and summarization can each run on a different model — and Ollama support means you can run entirely offline.

Do I need API keys?

You can use API keys from any supported provider, a Claude subscription via the Claude CLI, or no key at all with local models through Ollama. Keys are stored encrypted on your machine and are never sent anywhere except the provider they belong to.

Does it run on macOS?

Not currently — OneStone ships for Linux (AppImage, .deb) and Windows (installer and portable).

What can the agent actually do?

105+ built-in tools: web reading and search, file and shell access, email via AgentMail, calendar, contacts, scheduled routines, persistent memory, browser automation, and analytics — plus any external MCP server you connect.

Download OneStone

Free to use. Available now for Linux and Windows.

Windows

Setup · Portable

Requires 64-bit OS. Windows 10+, Ubuntu 20.04+. All releases on GitHub

Local-first. Your data never leaves.

Your conversations, memory, routines, and configuration live in ~/.onestone/ on your machine. No account. No telemetry. No lock-in. Export everything, anytime.