Ori ships with 22+ built-in tools. The Skill Graph automatically picks the right ones — you never need to specify which tool to use.
File system
| Tool | What it does | Approval |
|---|
| read | Read file contents or list directories | Auto |
| write | Create or modify files (shows diff preview) | Required |
| listDirectory | Browse directory trees with sizes and types | Auto |
Shell
| Tool | What it does | Approval |
|---|
| runCommand | Execute terminal commands | Always required |
Shell commands always require approval — even in trusted mode. Ori shows you the exact command before running it.
Web
| Tool | What it does | Approval |
|---|
| webSearch | Search the web via DuckDuckGo | Auto |
| crawl | Extract readable content from a URL | Auto |
Computer
| Tool | What it does | Approval |
|---|
| screenshot | Capture your screen | Auto |
| click | Click at screen coordinates | Auto |
| type | Type text into focused fields | Auto |
| scroll | Scroll the active window | Auto |
| pressKey | Press keyboard shortcuts with modifiers | Auto |
| mouseMove | Move the cursor to coordinates | Auto |
| getActiveWindow | Get app name, window title, and bounds | Auto |
System
| Tool | What it does | Approval |
|---|
| systemInfo | OS, CPU, memory, and environment details | Auto |
- A tool approval card appears showing what the tool wants to do
- You click Approve or Reject
- If approved, the tool runs and Ori continues
Session trust
When you approve a tool for a path, Ori remembers for the session:
- Trust is path-scoped — approving one directory doesn’t trust all directories
- Trust has a 30-minute TTL — automatically expires
runCommand is never auto-trusted — every command needs explicit approval
Self-correction
Ori doesn’t just call tools — it interprets results and self-corrects:
- If
read("/src/index.ts") returns “file not found”
- Ori tries
read("/src/") to list the directory
- Discovers the file is
index.tsx (not .ts)
- Reads the correct file automatically
Up to 3 self-correction attempts per tool call.
Extending with MCP
Ori supports the Model Context Protocol (MCP) for external tools. When you connect a service via Connectors, its tools are automatically added to the Skill Graph. 
