Tools: What the AI Can Do

The AI doesn't just chat—it can take real actions. Use tools to browse the web, read documents, generate images, manage your calendar, and more. The AI automatically decides which tools to use based on what you ask for.

Caiioo uses a read/write access model: Free users get read-only access to most tools, while Pro unlocks full write access across the board.

Everyone Gets These Tools (Free)

Web Browsing (Read-Only)

The AI can navigate your browser, read pages, take screenshots, search Google, and extract content. Perfect for research and data gathering.

What you can ask:

  • "Read this page and summarize it"
  • "Take a screenshot of this"
  • "Find all the prices on this page"
  • "Search Google for the best camping tents"

Google Workspace (Read-Only)

Search and read your Gmail, Google Drive, Docs, Sheets, and Calendar — no setup beyond connecting your Google account. Gmail searches understand natural filters—sender, subject, label, category, age, attachments, and unread state—so you don't have to know Gmail's search syntax.

What you can ask:

  • "Search my Drive for the Q4 report"
  • "Find unread emails from Bob in the last week that have attachments"
  • "Read my latest emails"
  • "What's on my calendar today?"

Sundial Agenda (Read-Only)

View your calendar events and reminders, check availability, and find free time slots. Works with Google Calendar on all platforms.

Web Search

AI-powered search with citations. Ask questions and get sourced answers instead of hunting through search results yourself.

Slate Editor

Real-time AI collaboration for code and documents. See Slate for details.

Workspace Files

Point Caiioo at folders on your computer and it reads, searches, and edits your files right where they live — nothing is copied into the app, so you never end up with two versions of a document drifting apart. Works on macOS, Windows, and Linux, and in the browser extension. (iPhone and Android have no folder access, so it's unavailable there.)

Adding folders: Open the + menu next to the message box and choose Add folder… — your system's normal folder dialog opens (the browser extension uses the browser's own folder picker). Folders are set up once; each conversation then picks which of them it can see with checkboxes in that same menu, and a new conversation starts with none active until you check one. There's also Browse folder…, which drops a file into your message as a lightweight reference — Caiioo opens it from disk only when it actually needs it.

Built-in safety: Before overwriting a file, Caiioo keeps a one-step undo. It refuses to replace a Word, Excel, or PDF document with a plain-text version of itself. Removing a folder from Caiioo never deletes anything on disk. And if the work leads somewhere you haven't shared, the agent can ask for a specific folder by path — you see the exact folder in an approval prompt, and saying yes grants access for that conversation only.

What it can do: Read PDFs and Word documents directly, search with the surrounding lines included so matches arrive with context, make precise multi-part edits that either all apply or leave the file untouched, show you exactly what a proposed rewrite would change before writing anything, and copy files between folders without altering them.

Run real commands (desktop, off by default): On macOS, Windows, and Linux you can also let Caiioo run shell commands — npm test, a build, a script — starting in the right project folder and confined to a sandbox that can't reach the rest of your disk (or the network, unless you allow it for that folder). Long-running commands like a dev server or a full test suite can run in the background while Caiioo carries on with other work. A Watch output control lets you follow the live output yourself, and the AI can ask to be woken when a job finishes — it comes back with the exit status and reads the stored output itself. The complete output is kept, so the AI can search or re-read it later instead of running the command again. Turn these on in Settings > Tools, Modes & Apps > Tool Configuration > Workspace Files.

Code Intelligence

Caiioo can understand your code the way a developer's editor does, instead of just matching text. Searching for a name finds everything that happens to share it — the definition, unrelated things with the same name, and mentions in comments alike. Code Intelligence asks a real language server instead, so Caiioo can jump to where something is defined, find the places it's genuinely used, read a symbol's type and documentation, list what a file defines, and see the errors and warnings your editor would show. Works on macOS, Windows, and Linux, in the folders you've shared with Caiioo.

Calculator

Quick math. The AI can do arithmetic, trigonometry, statistics, and more without using external tools.

API Integration (Read-Only)

Make GET requests to external REST APIs. Useful for fetching data from services we don't have built-in support for yet.

SQL Database

Create and query local SQLite databases. Useful for analyzing CSV data, building lightweight dashboards, or prototyping data workflows.

Sub-Agents

Delegate parts of a complex task to parallel agents so they run independently and report back. Each helper gets its own work budget and runs until its work is done, it stops making progress, or the budget runs out — then reports back what it has. This is useful when you want research, analysis, and drafting to happen at the same time. Sub-agent results render inline in the main chat.

Hearing

Ask follow-up questions about any audio attachment. The assistant can re-listen to a recording with a targeted question — "which words were mispronounced?", "what's the tone of voice?" — on the turn you attach it or any later one, via an audio-capable helper model. See Voice for the recording workflow.

Ask User

Pause an AI run mid-execution and surface a decision dialog. The AI presents up to 4 options (approve, approve with notes, reject, reject with notes) and waits for your input. Your notes flow back as plain-English guidance that overrides the plan, and the agent continues in place.

Self Checker

Rate and verify every assistant turn. Click the ⚖ button next to any response to open a verdict card. Choose from LLM-powered judgment plus deterministic checks (exact match, contains substring, regex pattern, number range, arithmetic). The judge can also write a small Python program to re-verify things no simple check can — recomputing a claimed statistic from the source data, for example — run locally in the same no-network sandbox as everything else (browser extension and desktop apps); a check that can't run cleanly reports as failed rather than silently passing. The verdict computes in the background, so it completes even if you close the panel. Results show inline. Cost rolls into your thread total. You can pick which provider and model act as the judge — leave the model blank and Caiioo picks one that suits the provider you chose.

Instant Tool Chooser

On-device semantic tool selection. The AI picks the right tool in ~10ms without calling a model. Enabled by default on every tier—toggle "Instant Tool Chooser" vs "Quick Tasks LLM" in Settings > Tools, Modes & Apps > Tool Configuration.

Pro Tier Tools ($9/month)

Pro unlocks full write access to tools that are read-only on Free, plus additional capabilities.

When you're running a local chat model (like Ollama), AI-powered tools that would send your data to a remote AI provider ask for your approval first. See Privacy & Data → Remote AI Providers.

Full Web Automation

Everything in read-only browsing, plus: click links, fill forms, type text, interact with page elements, and execute JavaScript. Perfect for form-filling, data entry, and browser automation.

What you can ask:

  • "Fill out this form with my info"
  • "Click on the Reviews tab and read what people say"
  • "Log into this site and download my invoice"

Full Google Workspace

Create, edit, and manage Google Docs, Sheets, Slides, Gmail drafts, Drive files, and Calendar events. Google Sheets gained new actions: paste CSV/TSV/HTML, split text to columns, trim whitespace, remove duplicates, apply and clear toolbar filters, move rows and columns, insert and delete cell ranges, protect ranges, define and update named ranges, attach developer metadata, and apply conditional formatting. Google Docs gained native comments (add, reply, resolve, delete), multi-tab document support, and smart-chip recognition (people, links, equations, page breaks, dates).

Google Drive — Create folders, move files, manage sharing permissions Gmail — Draft and send emails, download attachments Google Docs & Sheets — Create and edit documents, write formulas, format cells Google Slides — Create presentations, add text/images/tables, edit layouts Google Calendar — Create events and reminders, schedule meetings across every calendar in your account (team, family, and personal calendars)

Full Sundial Agenda

Create events and reminders, schedule meetings, and manage your calendar across providers.

Full API Integration

POST, PUT, PATCH, and DELETE requests to any REST endpoint — not just GET.

Document Ingestion

Upload and analyze PDFs, Word docs (DOCX), Excel spreadsheets (XLSX), and images with OCR. Higher-quality extraction on complex documents is available through OCR models accessed via your OpenRouter key.

Image Generation

Create images from text descriptions. Available models include FLUX.2 (Flex, Klein, Max, Pro), Gemini (2.5 Flash, 3 Pro, 3.1 Flash), GPT-5 Image, Seedream 4.5, and Riverflow v2. Perfect for illustrations, mockups, or visualizations.

What you can create:

  • A product mockup for a new design
  • An illustration for a blog post
  • A texture for a 3D project
  • Variations on an existing image

Video Generation

Generate videos from text descriptions. Available models accessed via OpenRouter: Google Veo 3.1, OpenAI Sora 2 Pro, and ByteDance Seedance. Valid durations and resolutions vary per model. Videos save as thread attachments.

What you can generate:

  • Product demo videos
  • Animated explainers
  • Scene transitions for edits
  • Storyboard sequences

Music Generation

Generate original music from text descriptions via Google Lyria 3 Pro Preview (accessed through OpenRouter). Creates royalty-free tracks that save as inline audio attachments in your thread.

What you can generate:

  • Background music for videos
  • Ambient soundscapes
  • Musical themes for projects
  • Instrumental versions of descriptions

Seeing-Eye Dog

Vision fallback for text-only models. If your chosen model doesn't support images (like local Ollama, DeepSeek V4 Pro, Kimi K2.6), attach images anyway—they route through a cheap vision model to generate captions, then text is sent to your main model. Default is Gemini 3.5 Flash Lite via OpenRouter. Auto-caption happens at message-build time with per-attachment caching. Use the vision({action: "inspect"}) tool for targeted follow-up.

Test Runner

Run a list of prompts with graders — substring match, pattern (regex), expected tool calls, or second-model 1-10 scoring. Each prompt runs in its own fresh conversation, through the same pipeline as your actual messages — so results reflect what the app really does. A floating panel shows the live run (suite name, per-test pass/fail, progress) with a cancel button, and results are saved as the run progresses, so exports and reports survive an app restart. Export results as a CSV pass/fail report.

XLSX Cell-Level Tracked Changes

Slate spreadsheets now support DOCX-style redlining via propose_change({editMode: 'xlsx_cell'}). Changes are anchored by cellRef and sheet name, rendered inline as <del>old</del><ins>new</ins>, with a toolbar for next, previous, accept, and reject. AI and user merge cell-by-cell with user edits winning on conflict.

Physics & Structural Analysis

Calculate projectile motion, collisions, energy, momentum, force, impulse, velocity-to-target, beam loading, column buckling, and material properties.

Private Sync

Sync your settings and conversations across devices via encrypted cloud backup. See Private Sync for details.

MCP Servers

Connect to remote MCP servers over HTTP/SSE, or run desktop tools (local MCP via the desktop app) on your own machine. Build custom tool integrations or connect to third-party services.

Meeting Recall

Retrieve details from recent video calls — transcripts, action items, and summaries so you can pull meeting context into any conversation.

Messaging Gateway

Answer and send messages across WhatsApp, Telegram, and more, with the agent responding for you. See Messaging Gateway.

More Pro Features

  • Unlimited custom modes — Create your own AI personalities with custom prompts and variables
  • Custom profile variables — Personalize AI behavior across all modes
  • Per-action instructions — Customize how each tool action behaves
  • Caiioo Benchmarks — Compare how models perform with quality evaluations and throughput tests
  • Priority support — Submit support tickets directly from the app

Experimental Tools

Toggle experimental tools via an on-device switch. These rotate as features mature into Free or Pro tiers. Available options include GitHub integration, Slack, advanced spatial reasoning, test automation, and more.

Enable or Disable Tools

Go to Settings > Tools, Modes & Apps > Tool Configuration to see what's available and toggle tools on or off. Some modes come with specific tools pre-configured.

Tools that don't run on your current device appear grayed out with a note saying which platforms they work on — and a tool that just needs the desktop app running says so — so you can discover what's available elsewhere instead of never knowing it exists. If you ask the assistant for one by name, it tells you which platforms the tool works on instead of claiming it doesn't exist.

See Also

This guide is maintained by the Caiioo team using Slate, our built-in editor.