What can your agent do?

Short answer: a lot. Longer answer below, organized by “the thing you’re probably trying to do.”

Email

Your agent can connect to your Gmail, Outlook, iCloud, Yahoo, or any IMAP inbox and then:

Summarize your inbox — “What happened overnight? What do I actually need to respond to?”
Draft replies for your review (or send directly, if you trust it that much)
Search your history — “When did I last email Sarah about the Q2 report?”
Label, archive, and organize — applying the rules you describe, not rigid filters
Flag urgent items and follow-ups that are slipping

Calendar

Connect Google Calendar or iCloud and your agent can:

Tell you what’s on your calendar today/this week
Find time for a meeting across multiple people’s calendars
Create events and invitations
Block focus time based on your energy patterns
Warn you about conflicts before they happen

Through Post for Me, one connection gives your agent publishing access to X, LinkedIn, Facebook, Instagram, Pinterest, and TikTok. It can:

Draft platform-optimized posts (LinkedIn is long, X is short, Instagram needs visuals)
Schedule posts for best times
Post with your approval, or fully automatically if you trust it
Monitor engagement and flag replies that need your attention

Browser — the headline capability

This is the one that makes the platform different. Your agent drives a real web browser — not a scraper, not a stripped-down headless client, a genuine Chromium instance that can click buttons, fill forms, log in, and see the page the way you do.

Two modes work in parallel:

Built-in sandbox browser (Camoufox). A stealth Firefox fork that runs inside the agent’s container, with randomized fingerprints, residential-style behavior, and zero credit cost. Good for: research, public pages, scraping, data collection, anything that doesn’t need to be logged in as you.
Browser Control extension for your own Chrome. Install the one-click extension on your laptop and your agent can drive your personal browser — with your logged-in sessions, your saved passwords, your cookies, your everything. Good for: anything inside a gated dashboard (GitHub, Stripe, Slack web, HubSpot, your bank, your email provider’s web UI), personal SaaS tools, sites that require SSO, and anywhere a scraper would trip a bot check.

What that unlocks in practice:

Deep research — real multi-hop browsing across sites, not just search snippet summarization. It opens pages, reads them, cross-references them, writes you a briefing.
Monitoring — check your competitors’ pricing pages, hiring pages, or changelogs on a schedule and surface diffs.
“Go do this in my browser” — “Open my Stripe dashboard, pull up this month’s revenue chart, and screenshot it for me.” “Log into GitHub notifications and triage the ones I haven’t read.” “Check my Linear backlog for anything tagged blocker.”
Form filling and SaaS workflows — any web app you operate today, your agent can operate with you watching. Click, type, navigate, submit.

Browsing is free on every tier. Both the sandbox and the extension cost zero token credits. Load as many pages as you want. Only the LLM thinking about the page content uses your included AI budget.

There’s a full Browser Control walkthrough covering install, pairing, security model, and what your agent can and can’t do in your real browser.

Scheduled tasks

Anything your agent can do once, it can do on a schedule. Examples:

“Every morning at 7am, send me a summary of my inbox and calendar”
“Every Monday, check our three top competitors and tell me what changed”
“Every Friday afternoon, draft a weekly update post for LinkedIn”
“Every first of the month, calculate my credit usage and warn me if I’m trending over budget”

Trial tier gets 5 cron jobs. Standard gets 15. Pro gets 50.

Memory and learning

Your agent keeps two files it can read and write:

MEMORY.md — environment facts, lessons learned, preferences
USER.md — its understanding of YOU, your role, your work style, your goals

These files are injected into its brain at the start of every conversation. That’s why it remembers what you told it last week — not because of some “session” or “history” trick, but because it literally has notes about you.

It also writes its own skill documents when it figures out a good way to do something. The next time a similar task comes up, it reads its own notes and executes faster. It gets better at working with you over time.

Chatting via your favorite channel

You don’t have to sit in the dashboard. Your agent can be reached via:

The dashboard chat
Telegram
Discord
Slack

All four share the same agent. Context and memory are continuous across channels — start a conversation on your laptop, pick it up on your phone, finish it on Slack.

Code and data analysis

Your agent runs in its own Linux container with Python, pandas, matplotlib, and standard data tools. You can:

Upload a CSV and ask it to analyze / chart it
Ask it to run scripts against data you paste in
Have it automate small tasks that are cheaper as code than as API calls

Approval controls

For everything above that takes action — sending email, posting to socials, deleting files, making purchases — you decide how much oversight you want. Three trust modes:

Require approval — nothing happens until you click approve
Auto + notify — it acts immediately but pings you
Full auto — it just does it and you review later

You set this per action type. Start cautious, loosen as you get comfortable.

What it can’t do (yet)

Being honest about limits:

It can’t make phone calls. Voice is on the roadmap, not shipped.
It can’t physically sign documents. DocuSign integration is planned.
It can’t access APIs that require custom authentication. If a tool needs a weird OAuth flow we haven’t built, you’re stuck.
It doesn’t have perfect recall of long conversations. Memory is curated, not exhaustive — it keeps what matters, not every word.
It’s not a replacement for human judgment on critical decisions. It’s an assistant, not a CEO.