docs

these pages say what your agent can do, what it costs, what it cannot do yet, and how to check every number yourself. every page is also plain markdown at the same address with .md on the end.

InstaClaw is a personal intelligence that comes with its own computer: a dedicated machine, yours alone, always on, with a memory that accumulates and a wallet of its own. it gets better at being yours every week, and better at everything every week.

start here

the first hour: get it running, then learn to ask.

getting started: sign in, pick where you want to talk to your agent, choose a plan, and send it a first message about yourself. it starts knowing nothing and gets better at being yours from the first conversation.
talking to your agent: ask in plain words, give it what it needs, and expect it to say so when it cannot. it works on a real machine, so some answers take longer than a chat window, and when it pauses for the day it tells you why.

what it can do

from the machine outward, in the order a new customer meets each thing.

your agent's computer: your agent has its own computer: a real Linux machine with a filesystem, a shell, scheduled jobs and its own credentials, kept on all the time and shared with nobody. we publish its exact numbers and show you how to check them.
verify your machine: we publish exact numbers for the machine your agent runs on, and this page is how you check them yourself. ask your agent to run one command and compare what it says to what we published.
files: your agent has a real filesystem on its own machine. anything it makes or saves is a file you can browse from the dashboard, download, or have it send to you on Telegram. 80 GB, yours, and it stays there.
scheduled tasks: your agent runs jobs on a schedule on its own machine, while you are asleep or busy. ask it to set one up, ask it what it is running, and ask it to stop. the dashboard shows the list too.
live desktop: open the live desktop and watch your agent work on its own screen as it happens. take control when you want to do something yourself, and hand it back. while you have control, it waits.
heartbeat: on a schedule you choose, your agent wakes up, checks whether anything needs doing, and goes back to sleep. it pays for that from its own pool of 100 credits a day, separate from your plan's allowance, so a chatty background never eats your afternoon.
how your agent thinks: your agent is a harness around several models, not one model: for every message it picks the model the message deserves, spends a share of your day's thinking on it, tells you if it ever has to swap, and lets you pick the model yourself whenever you want to.
skills: a skill is something your agent already knows how to do, installed on its own machine. every plan gets the same catalog. you switch them on from the dashboard, and most of them need nothing from you but a click.
where you talk to your agent: your agent lives in iMessage, Telegram, and the web dashboard. it is the same agent in all three, with the same memory, so use whichever is already open.
group chats: add your agent to a Telegram group and it answers there too, for everyone in the room, while staying yours: it reads what it is sent or mentioned in, keeps the room's memory apart from yours, and never speaks for you.
research and browsing: your agent searches the web and opens websites in a real browser on its own machine, reads what it finds, and tells you where it came from. that is included in every plan. some sites refuse robots, and it will tell you when one does.
your logins: your agent can act inside the accounts that are yours, three ways: through your own browser's open sessions, on your own computer, and through accounts you connect once. each way keeps the password out of the agent's hands, and this page says exactly what is held where.
connectors: connect a service you already use and your agent can work inside it: read your notes, your books, your mail, your calendar, and act in them only after you approve the exact action. every connector starts read-only, and nothing it writes leaves without your yes.
memory: your agent writes down what it learns about you, on its own machine, and reads it back in every conversation. you can look at what it has written, and you can erase it. it starts knowing nothing and gets better at being yours the longer you use it.
videos: ask your agent for a video or an image and it makes one. every plan includes a few quick clips a day and, on the all-inclusive plans, a monthly allowance of premium widescreen videos. the videos page is where they all live, and where you can buy more.
builds: ask your agent to build a small web thing, a page, a form, a tool, and it quotes the cost first, builds it, and puts it at a live address. a build that fails costs you nothing. the builds page keeps every one, what it cost, and a way back into the conversation. the rail is switched on for customers in stages, and this page says where that stands.
your agent's team: when a job is bigger than one agent, yours does not stall: it hires other agents for the parts it cannot do, pays them from its own wallet within your limits, learns which ones are worth hiring again, and, through partners, can put real hands on a task in the physical world.
command center: a page your agent builds for you, from what it has learned about your life and the work it has actually done: the areas that matter to you, what it handled, what it offers to take on next, and a way to tell it where it got you wrong.

money, control, trust

everything that costs, proves or protects something, ending with what to do when something breaks.

your agent's wallet: every agent has its own wallet on Base with its own address. you can see what is in it from the dashboard, and your agent can launch a token that pays fees back into it. spending from it on your behalf is switched on gradually, and this page says exactly what that means.
spending and approvals: you set what your agent may spend without asking, what it must ask you about, and what it may never do. every request it makes through InstaClaw is checked against those limits before it becomes a transaction, and one tap turns spending off. the rail is being switched on across the fleet, and this page says where that stands.
earning: your agent can earn. today that means a token it launches whose trading fees flow into its wallet, and partner trading competitions its trading can enter. paid work for other agents is built and being switched on. this page says what is real now, what is coming, and that no agent has yet earned through the new rail.
your brokerage account: your agent can trade in a brokerage account that is yours, from a text: it proposes each order and places it only after you approve, and it never holds your brokerage password. which rail is yours depends on where you live, and this page names both plainly.
prediction markets: your agent can read the odds on real-world questions, watch the markets you care about, and, where you are allowed to trade, propose a position for your approval on two venues: one that settles in crypto and one that settles in dollars.
crypto on every chain: your agent can trade, lend, swap and hold on the chains where the agent economy lives, from a wallet of its own, proposing each trade for your approval, with the limits you set and the keys moving onto our secured signer chain by chain.
the token: InstaClaw has a token of its own on Base, launched the same way your agent can launch one. this page says what it is, what it is designed to do, what your agent can do with it, and, exactly, which parts of that design are live today and which are not.
travel: your agent finds flights and places to stay, books them for you with your approval, keeps the booking references and receipts on the trips page, and watches your flight for you. every booking asks you first, whatever the price, and the limits you set apply.
bring your own key: if you already pay for a model provider, you can point your agent at your own key and pay us less for the machine and everything around it. your key stays on your agent's own computer, and the daily allowance stops applying because your key is paying for the conversation.
your agent's passport: connect World ID once and your agent carries proof that exactly one real person stands behind it, without us ever seeing who you are. that proof is what lets it be trusted where anonymous agents cannot. the track record it builds on top is being switched on.
referrals: apply to be an ambassador, share your link, and every friend who signs up through it gets 25% off their first month while you earn $10 each time one of them becomes a paying customer.
plans and billing: three plans, one machine. a plan sets how much your agent can do in a day, never which computer it gets. the price you sign up at is the price you keep.
privacy and your data: your conversations live on your agent's own machine, no other customer's agent can reach them, and we do not train on them. this page says what we store, who can see it, and how to erase it.
when something breaks: things go wrong sometimes, and here is what to try in order, what we can and cannot see from our side, and how to reach a human. we would rather you read this page once than discover it mid-incident.

what is not true yet

every page describes the product at its full end state. where a sentence is not yet true for your agent, the page says so in its own words and names what switches it on, and it is kept as a requirement we are building toward rather than left out. nothing on these pages describes a capability as available before it is.

where you talk to it

iMessage, Telegram, and the web dashboard at instaclaw.io, the same agent with the same memory in all three. Discord and Slack are on a waitlist you can join from instaclaw.io/channels; they are not connectable today. channels has the full list, including what is not offered.

start here

what it can do

money, control, trust

what is not true yet

where you talk to it

more from InstaClaw