Built-in Tools Reference

Every option for the tools Pando ships with: reading web pages, searching the web, driving a browser, driving your desktop and converting documents. For the explanations see Browser Automation, Desktop Controller and Document Conversion; for the step-by-step see the guide Give Pando eyes and hands.

All keys live in the [InternalTools] section of .pando.toml.

Web reading and search

[InternalTools]
FetchEnabled            = true
FetchMaxSizeMB          = 10
GoogleSearchEnabled     = true
GoogleAPIKey            = ''
GoogleSearchEngineID    = ''
BraveSearchEnabled      = true
BraveAPIKey             = ''
PerplexitySearchEnabled = true
PerplexityAPIKey        = ''
ExaSearchEnabled        = false
ExaAPIKey               = ''
SourcegraphEnabled      = false
SourcegraphToken        = ''      # optional: the public service is used when empty
Context7Enabled         = true    # library documentation, no key needed

The fetch tool can read a page through a browser when the page needs JavaScript:

{
  "url": "https://example.com",
  "format": "markdown",
  "browser": "auto"
}

browser accepts auto, chrome, firefox, curl and http.

Browser

[InternalTools]
BrowserEnabled     = true
BrowserType        = 'chrome'
BrowserExecutable  = ''         # empty: find it automatically
BrowserHeadless    = false      # true: no visible window
BrowserTimeout     = 30         # seconds
BrowserUserDataDir = ''
BrowserMaxSessions = 3

`BrowserType`	Browser
`chrome`	Google Chrome (default)
`msedge`	Microsoft Edge
`chromium`	Chromium
`opera`	Opera
`firefox`	Firefox (through the fetch tool)
`lightpanda`	Lightpanda, a light browser with no window
`obscura`	Obscura, a fast browser with no window, written in Rust

Lightpanda and Obscura are started by Pando as a background program and have no window and no user profile, so BrowserHeadless and BrowserUserDataDir do not apply to them. For Obscura, the obscura command must be in your PATH.

Browser sessions are shared from a pool limited by BrowserMaxSessions. If your normal browser profile is in use, Pando falls back to a temporary one.

Tool	What the agent does with it
`browser_navigate`	Open an address
`browser_get_content`	Read the page: HTML, text or title
`browser_screenshot`	Take a picture of the page or of one element
`browser_click`	Click something
`browser_fill`	Fill a form field
`browser_scroll`	Scroll
`browser_evaluate`	Run JavaScript in the page
`browser_console_logs`	Read the page’s console messages
`browser_network`	See the requests the page made
`browser_pdf`	Save the page as PDF

Desktop controller

[InternalTools]
DesktopEnabled            = false   # master switch
DesktopBackend            = 'auto'  # auto | atspi | uia | ax | cdp | null
DesktopAllowPhysicalInput = true    # allow a real mouse click or key press as fallback
DesktopMaxNodes           = 500     # most elements reported per look
DesktopDefaultDepth       = 3       # how deep to look into a window by default
DesktopActionTimeout      = 10      # seconds
DesktopSnapshotTTL        = 60      # seconds a look stays usable
DesktopScreenshotScale    = 1.0     # shrink screenshots before sending them to the model
DesktopAllowedApps        = []      # if set, only these apps can be touched
DesktopDeniedApps         = []      # never touched; wins over the allow list

Key	What it is for
`DesktopBackend`	Leave on `auto`. Pando picks the right one for your system and, when a browser session is open, drives the browser through the same tools. Any other value pins that one
`DesktopAllowPhysicalInput`	When an app does not offer a proper way to act on an element, Pando can fall back to a real click or key press. `false` forbids that
`DesktopAllowedApps` / `DesktopDeniedApps`	The practical way to fence the agent in: allow only the app you are working on, deny your password manager, mail or terminal. Deny always wins
`DesktopScreenshotScale`	Lower it (for example `0.5`) to make screenshots cheaper

What the agent can do	Asks you first?
List running apps and their windows	no
Read a window’s content and structure	no
Find an element by name or role	no
Wait for something to appear, disappear, become enabled or focused	no
Click, focus, type text, press a key or key combination, scroll	yes
Take a screenshot of the screen, a window or one element	yes
Click a raw screen position (last resort, by sight)	yes

Platform	What you need
Linux (X11)	An accessibility bus running (`org.a11y.Bus`); GNOME and KDE have it on by default
Linux (Wayland)	Say yes to the desktop dialog the first time. Your answer is remembered
macOS	Give Pando Accessibility permission (System Settings → Privacy & Security → Accessibility). Screenshots also need Screen Recording
Windows	Nothing extra

The desktop tools can also be offered to other programs through pando mcp-server. They follow the same DesktopEnabled switch.

Document conversion

No options. It works out of the box.

pando convert report.docx              # print the Markdown
pando convert data.xlsx -o data.md     # write it to a file
pando convert https://example.com/page.html
pando convert --list-formats

Kind	Extensions
PDF	`.pdf`
Word	`.docx`
Excel	`.xlsx`, `.xls`
PowerPoint	`.pptx`
Web	`.html`, `.htm`
Data	`.csv`
E-book	`.epub`
Notebook	`.ipynb`
Feed	`.rss`, `.atom`
Markup	`.xml`, `.json`, `.jsonl`
Archive	`.zip`
Text	`.txt`, `.md`, `.markdown`

To add a whole folder of documents to the knowledge base, converting on the way:

pando kb import /path/to/documents