Is it actually private?
Yes. CtxAnt runs entirely on your machine. The Chrome extension talks to the CtxAnt app over ws://127.0.0.1 — nothing goes to a CtxAnt-owned server, because there is no CtxAnt-owned server. Your page content is only sent to the AI provider you chose (xAI or Anthropic), authenticated with your API key. Full privacy policy.
What does it cost?
BYOK — you pay the AI provider directly, not us. On xAI Grok the average user spends under $2/month running 5 macros/day. If you already pay for Claude, it works on that too. No CtxAnt subscription, no trial expiration, no credit card.
Can it see my banking / medical / personal stuff?
Only when you ask it to. CtxAnt only acts on a tab when a command comes in from your own Telegram bot. The upside of that is: it's already logged into your banking, so the answer to "what did I spend this month?" takes 3 seconds and a screenshot, not a finance migration. Data never leaves your machine except via your own AI provider's API.
Why Telegram?
It's the only chat app with a free, programmable bot layer, no approval process, and instant cross-device delivery. You already have Telegram open. You already know how to use it. And because every agent gets its own bot, deploying a new agent feels like hiring a coworker — their messages show up in their own chat bubble in your sidebar.
Why BYOK instead of charging per-use?
Three reasons. (1) Your data stays between you and your AI provider — we never see a prompt. (2) Costs scale with your usage, not our pricing tiers. (3) We don't have to touch credit cards, which means we can ship faster and can't be held hostage by Stripe.
Windows? Linux? iPhone?
Mac first. Windows and Linux are coming — most of the code is portable, but the installer, menu-bar UI, and onboarding are Mac-specific. Join the waitlist on the repo. iPhone: not planned — iOS can't drive Safari the way Chrome Extensions drive Chrome, which is the whole point.
How is this different from Claude "Computer Use" or ChatGPT Operator?
Those run the AI in their sandbox, controlling their browser. They don't know your Gmail is already open or that you're logged into LinkedIn, so every task starts with 20 seconds of authentication theatre. CtxAnt skips all that because the browser is yours. You never paste a password into the AI, you never share a session, you never wait on a fresh login.
What if the AI does something dumb?
You're watching a live Chrome window — so you see everything. For destructive actions we can require explicit approval before submit; that's configurable per-agent. And the menu bar always has a one-click Stop.
Is it open source?
Yes. Code on GitHub. We'd rather you read the ten files that run on your machine than take our word for it.
