Getting started
Your Gina AI agent is an AI worker you talk to in plain language. It researches, writes, works in the apps you connect, runs jobs on a schedule, and, if we host an app for you, builds and changes that app. This page is the map: the words we use, what is yours, what you can build, and where to read more.
Setting up for the first time? Start with Set up your agent. You can also ask your agent any question on this page: it knows these docs and will link you to the right one.
What you have
We run it for you
Your agent
- Its own private server, which we operate, update and back up.
- Its memory, your threads, its files and its automations.
- The credentials and connections you give it, stored encrypted.
- You reach it in the console, in Slack, or from other tools with an API key.
Yours, managed for you
Your hosted app (optional)
- A web app for your business: CRM, Client portal, or a blank slate.
- A cloud account that belongs to your business, holding only your app and its data.
- A test site for trying changes and a live site for your users.
- Your agent changes it for you; the live site changes only after an approval.
The split matters when you think about ownership. The agent is a service we run for you. The app, its code and its data belong to your business, and live in an account kept apart from every other customer's and from our own systems. The agent reaches into your app's account only to do the work you ask for, with short-lived access, and it never changes your live site on its own.
Words we use
- Agent
- Your AI worker. It has its own private server, memory, credentials and automations, and does work you ask for in chat or on a schedule. More
- Console
- Where you work with an agent: Threads, Canvas, Automations and Files on the left, and the Menu for everything else. More
- Thread
- One conversation with your agent. Private to you, shared with specific people, or open to everyone with access to the agent. More
- Task
- One piece of work the agent runs: a reply to your message, or one run of an automation. More
- Automation
- Work that runs by itself: on a schedule, after a calendar meeting, or when something changes. Also called a schedule. More
- Connection
- An account in another app that your agent can use for you, such as your email or calendar. More
- Credential
- A key or token your agent can use (your Claude account, GitHub, Slack, your own keys). Stored encrypted, never shown in chat. More
- Inbox
- Everything waiting on you or your team: approvals, questions, mentions and updates. More
- Approval
- A decision the agent will not make alone. It asks, stops, and carries on when you answer. More
- To-Do's
- The goals your agent is working on, each with a plan you can read and steer. More
- Canvas
- Your agent's dashboard. You can design new tabs with the agent. More
- Memory
- What your agent keeps between tasks: your preferences, decisions and facts about your business. Say "remember that…". More
- Hosted app
- A web app that belongs to your business, in a cloud account created for you and managed by us. Your agent builds and changes it. More
- Test site
- Where every change to your app lands first, so you can try it before your users see it. Also called staging. More
- Live site
- The version of your app your users see. A change reaches it only after an approval. Also called production. More
- Deploy receipt
- The proof that a change is running: where, which version, and when. The agent never says a change is live without one. More
- App data
- The records inside your app (contacts, deals, clients). The agent brings data in through your app's own API, after you approve the batch. More
- Member and role
- Someone who can use your agent, as owner, admin, operator or viewer. More
- Agent API key
- Lets an AI tool (ChatGPT, Cursor, Claude, your own software) use one agent as you, with the access you choose. More
- Connector (MCP)
- Your agent added to Claude, ChatGPT, Cursor or another AI tool through its MCP server address, so you can use it from there. You sign in, or use an agent API key. More
- App client
- A client ID and secret you make on the Access page for an app that asks for one before it connects to your agent. Each person still signs in. More
- Webhook
- An address (an inbound endpoint) one of your apps calls when something happens. Your agent then follows the instruction you set for it. More
- Live stream
- The events your agent publishes, followed live or caught up on by your app or dashboard with a key you create, or pushed to your app's address. More
- Listen
- A reply read aloud, a point at a time, with speed and skip. More
- Voice message
- Dictation: speak instead of typing, for up to 20 minutes. Your agent gets the words cleaned up and the original transcript. More
- Credits
- Each agent uses one credit; hosting an app for it uses one more. More
Building blocks
Everything your agent does is made of a few blocks. Use one on its own, or combine them: the recipes in the next section show how. You ask in plain words; the examples are only a start.
Ask
Research, writing, analysis and answers, in a thread or in Slack.
“Compare these three venues and recommend one.”
Documents
Reports, PDFs and spreadsheets saved to Files, with a link you can share.
“Turn this into a one-page PDF for the board.”
Remember
Keeps your preferences and decisions for every future task.
“Remember that invoices go out on the 1st.”
Connect an app
Works in your email, calendar, documents and other apps you connect.
“Check my calendar for Thursday.”
Draft email
Drafts replies in your own mailbox, addressed and ready; sends only when you say so.
“Draft a reply to Dana's last email.”
On a schedule
Runs any task by itself, every day, week or hour.
“Every weekday at 8, send me what's on today.”
After a meeting
Runs a task when a calendar meeting ends.
“After each client call, write up the actions.”
When something changes
Checks something regularly and tells you when it changes.
“Tell me when a new review appears.”
Change your app
Builds a change to your hosted app, puts it on the test site, and takes it live on your approval.
“Add a birthday field to contacts.”
Bring data into your app
Loads records from a spreadsheet, file or shared link into your app, after you approve the batch.
“Here's my client list, add them to the CRM.”
Approve
Puts a checkpoint on anything that matters; the agent waits for your answer.
“Ask me before you send anything to a client.”
Dashboard
Adds a tab to your Canvas showing what you want to watch.
“Add a tab with my open deals and this week's calls.”
Goals
Tracks a bigger goal as a plan, step by step.
“Make a plan to move our clients off the old portal.”
From other tools
Lets Claude, ChatGPT, Cursor or your own software read your agent's work, send it tasks and answer its approvals, as a connector over MCP with a sign-in or a key.
“Create a key for Claude Desktop that can send tasks.”
When an app tells it
One of your apps calls your agent's webhook when something happens, and the agent follows the instruction you set for it.
“When the booking form sends a new booking, add it to my calendar and tell me in the morning.”
Tell your apps
Publishes an event when something happens in its work (a deal won, a report ready) for your apps to follow or receive.
“When you finish a client report, publish report.ready with the client's name and the link.”
Live feed
A dashboard or app follows what your agent publishes as it happens, or catches up later, with its own key; or each event is pushed to your app's address.
“Publish deal.won whenever a deal closes, so the sales dashboard shows it live.”
Listen
Reads a reply aloud, a point at a time, with speed and skip; on an iPhone through the loudspeaker.
“Keep the briefing short enough to listen to on the drive in.”
Dictate
Speak a message instead of typing it, for up to 20 minutes. Your words arrive cleaned up, with the original kept.
“Here's the brief for the spring launch, from the top…”
Put them together
The most useful work combines blocks: something happens, the agent acts in one of your apps, and you stay in control with an approval where it matters. Copy an ask, change the details, and send it.
New enquiry email becomes a lead in your CRM
Partly available“Every 15 minutes, check info@ for new enquiries. For each one, prepare a lead in my CRM with their name, email and what they asked about, and ask me before adding them.”
Each new enquiry arrives in your Inbox as a lead ready to add; one tap adds it to your app.
Morning briefing
“Every weekday at 7:30, read my calendar and inbox and send me the day's meetings, anything urgent, and three things to do first.”
A short briefing in its own thread every morning, and in Slack if you connect it.
After every client call, a recap and a follow-up
“After each meeting with someone from outside the company, write up the decisions and actions, remember them, and draft a follow-up email to the attendees.”
A recap in a thread and a follow-up waiting in your drafts, minutes after the call ends.
Load your records, then shape the app around them
“Here's my client spreadsheet. Import it, then add a view that groups clients by city.”
You approve the import, check the new view on the test site, and approve it for your live site.
A weekly report as a PDF
“Every Friday at 4, summarise this week's meetings, finished work and open questions in a one-page PDF.”
A PDF in Files every Friday, with a public link you can send to anyone.
A live status board
“Add a Canvas tab showing what's waiting on me, my open goals, and what the automations reported today.”
One tab that shows the state of everything the agent is doing for you.
Use your agent from Claude or ChatGPT (an API key and a connector)
“Create an API key for Claude Desktop that can send tasks and answer approvals.”
Add your agent as a connector in Claude, ChatGPT or Cursor (sign in, or paste a key from the Access page), then send it tasks, search its work and answer its approvals without opening the console.
An app tells your agent when something happens (a webhook)
Partly available“When the website form sends a new enquiry, look the person up in my inbox and draft a reply.”
Each enquiry starts a task at once, in its own thread, and a draft reply waits in your mailbox.
A live feed of what your agent does on your dashboard (a stream)
Partly available“Each time you finish a client report, publish report.ready with the client's name and the link.”
Your dashboard follows the stream with its own key and shows each report as it lands; after time offline (up to 7 days), it catches up from where it stopped.
Listen to replies while driving
“Every weekday at 7:30, send me a short briefing written to be heard: today's meetings and the three things that matter most. No tables.”
Press Listen before you set off: the briefing plays through your phone's loudspeaker, and you can speed it up or skip a point.
Everything else, in broad strokes
Each page below covers one part of the product in depth.
Working with your agent
- Threads and chat — Start a thread, keep it private or share it, attach files (iPhone photos and Outlook emails convert on their own), add to a task while it runs, leave a note the agent won't answer, and talk to it from Slack.
- Listen and dictate — Listen to replies read aloud, with speed and skip (on an iPhone through the loudspeaker), and dictation: speak a voice message of up to 20 minutes, your words cleaned up and the original transcript kept.
- Automations — Work that runs by itself: on a schedule, after a calendar meeting, or when something changes. Where results go, and when you're notified.
- Inbox and approvals — Everything waiting on you or your team in one list, and how to answer the agent's approvals: checks on your test site, going live, and imports.
- Connections — What your agent can use: the apps it signs in to for you (email, calendar, documents and more), how to connect or reconnect one, and how it drafts email.
- Webhooks and live events — Let your own apps and your agent tell each other when something happens: a webhook (inbound endpoint) an app calls, events your agent publishes, a live stream or catch-up feed for a dashboard, and push to your app's address. Being turned on agent by agent.
- To-Do's — The goals your agent is working on, each with its plan: what's in progress, what's next, and how to steer it.
- Files — Everything your agent produced, in one place: preview, search, favourite, and share files as public or private links. A private link opens only for the people you name, with a code emailed to them.
- Canvas — Your agent's dashboard: today's work, what's waiting on you, and tabs you design with the agent.
Your hosted app
- Your hosted web app — A web app in your own cloud account, managed for you: a test site and a live site, and approvals before any change goes live.
- Building with your agent — Best practices for changing your app with the agent: one change at a time, check it on the test site, keep a record of decisions, go live on your approval.
- Your app's data — Bring your records in (a spreadsheet, a file or a shared link), let people sign in to your app, and what the agent can and cannot change.
Account, team and help
- Access roles — Who can use your agent: share it with your team, what owners, admins, operators and viewers can do, and who may approve going live.
- API keys for other tools — What can use your agent: let ChatGPT, Cursor, Claude or your own software work with one agent over MCP, as a connector, with a sign-in or a key that has only the access you choose. App clients for apps that ask for a client ID and secret.
- Usage, account and plan — Usage and limits, your plan, your team, sign-in security, and preferences.
- Getting help — Book a session, report an issue, or ask your agent to get a person. What to include so we can help quickly.
- Security — How credentials are stored, how accounts are kept separate, how sign-in is protected, and how to report an issue.
Set-up guides
- Getting your credentials — What your agent can use: its Claude connection and the keys it works with (GitHub, Slack, your own), and where each one is stored.
- Slack app setup — Create a Slack app from a manifest, generate the bot and app-level tokens, and connect it to your agent.
- Finding a Slack member ID — How to look up the member ID for yourself or anyone in your workspace, for the agent's allowlist.
- Linking an AWS account — For agents that work in an AWS account you run yourself: a dedicated user, what it may do, a spending guardrail, and key rotation.
- Agent profiles — What each profile does: engineering, analytics, research, finance, marketing, and the minimal starting point.
- Agent state: codestate & memstate — Saved versions of your agent's code and memory, and how to switch between them.
Guides
- Building on AWS with an agent — Infrastructure as code, separate environments, least privilege and cost guardrails, for agents that build in your own AWS account.
- Autonomy, checkpoints, and stability — How much latitude to give an agent, where a human checkpoint pays for itself, and habits that keep scheduled work stable.
- Working with a system you cannot fully audit — Your agent will do more in a day than you can read in a day. How to keep real oversight anyway.
Plans and billing
- What you need first. A Claude Pro or Max plan of your own. Your agent thinks with it, so a free Claude account can't run an agent.
- Choosing a plan. Where plans are bought online, you choose your plan as part of setting up your agent, after you create your account. If your invite comes with an offer, it is applied for you at checkout.
- Paying securely. When you pay online, payment happens on a secure checkout page run by Stripe, our payment processor. We never see or store your full card number. Once your payment is confirmed, you carry on setting up your agent.
- The app included with some plans. If your plan includes an app we build and host for you, setup asks what it starts from: CRM, Client portal or Blank slate. You choose; your agent shapes it from there. See Your hosted web app.
- Changing or canceling. If you pay online, go to Account › Plan and press Manage billing. There you can cancel, update your card and see your invoices. A cancellation takes effect at the end of the period you have paid for, so your agent keeps working until then. To move to a different plan, email us at hello@talktogina.ai for now.
- If a payment fails. We let you know and give you time to update your card. If the payment still hasn't gone through when that time runs out, your agent pauses. Pausing deletes nothing: your agent's memory, threads and files are kept, and it starts again once the payment goes through.
Getting help
Ask your agent first: it can explain any feature and point you to the page. When you need a person, ask the agent to get support, or use Support to book a session or report an issue. See Getting help for what to include. What changed recently is in the release notes.