Core capabilities
Persistent memory
Ori remembers what you tell it — across every conversation, forever.- Automatically extracts facts, preferences, and patterns
- Organized into a tiered system: always-loaded summaries, detailed context on demand
- Toggle Recall on/off to control when memories are injected
- View, search, and delete memories in Settings
Computer use
Ori can see and control your screen — like a human sitting at your desk.| Action | What Ori does |
|---|---|
| Screenshot | Captures your entire screen and analyzes the content |
| Click | Clicks at specific coordinates on your screen |
| Type | Types text into any focused input field |
| Scroll | Scrolls up/down in the active window |
| Press keys | Presses keyboard shortcuts (⌘C, ⌘V, Enter, etc.) |
| Move mouse | Hovers over elements for tooltips or menus |
| Get active window | Identifies the current app, window title, and position |
Scheduled tasks
Set up recurring agent jobs that run on autopilot.- Morning briefings — “Summarize my GitHub PRs every morning at 9am”
- News digests — “Every evening, give me the top AI news”
- Monitoring — “Check my website every hour and alert me if it’s down”
- Runs even when Ori is in the background
- Results appear on your Home screen
Connectors
Connect Ori to your digital life via MCP (Model Context Protocol):| Connector | What it enables |
|---|---|
| GitHub | Browse repos, read issues, review PRs, search code |
| Gmail | Read, search, and draft emails |
| Google Calendar | View schedule, create events, check availability |
| Slack | Read channels, send messages, search conversations |
| Notion | Read pages, query databases, search workspace content |
Skill routing
Every request flows through a Skill Graph that picks the right tools automatically. You never need to specify which tool to use.- 22+ built-in tools organized into semantic domains
- Zero-token routing — no LLM calls needed for tool selection
- Adapts based on your usage patterns
Built-in tools
File system
- read — Read file contents or list directories
- write — Create or modify files with diff preview
- listDirectory — Browse directory trees
Shell
- runCommand — Execute terminal commands (always requires approval)
Web
- webSearch — Search the web via DuckDuckGo
- crawl — Extract readable content from any URL
Computer
- screenshot — Capture your screen
- click — Click at screen coordinates
- type — Type text into focused fields
- scroll — Scroll the active window
- pressKey — Press keyboard shortcuts
- mouseMove — Move the cursor
- getActiveWindow — Identify the current app
System
- systemInfo — OS, CPU, memory, and environment info
AI providers
Use any model from any provider — or run fully offline:| Provider | Popular models | Best for |
|---|---|---|
| Anthropic | Claude 4, Sonnet | Complex reasoning, coding |
| OpenAI | GPT-4.5, GPT-4o | General use, vision |
| Gemini 2.5 Pro/Flash | Large context, multimodal | |
| Ollama | Llama 3, Mistral, Phi | Full privacy, offline |
Platform features
Spotlight (⌥ Space)
Summon Ori from any application as a floating overlay. Ask a question, get an answer, dismiss — without switching windows.
System tray
Ori lives in your menu bar / system tray. Always running, always ready.Auto-updates
Updates download silently in the background. When ready, a non-intrusive prompt lets you restart to update.MCP server
Ori exposes its intelligence as an MCP server onlocalhost:4141. Connect from Cursor, VS Code, or any MCP-compatible tool.
→ MCP server reference
Privacy
- 100% local — all data stored in
~/.ori/ - Zero telemetry — no analytics, no tracking, no phone-home
- OS Keychain — API keys stored in macOS Keychain, Windows Credential Manager, or Linux Secret Service
- Your models — you choose the provider. Use Ollama for fully offline operation
- Open source — inspect every line of code