She talks
In chat, in voice and face to face, in one window. Every sentence names its source, and an answer without coverage in your knowledge never reaches the customer.
AI employee
Hire Ada. She talks to customers, books the meetings, presents on video calls and works in Teams and Slack.
Ada works on your website from the first question to a booked slot, then runs the demo herself, and joins the meeting with a face and a voice. Every answer rests on your company's knowledge and cites its sources. Every change to your data is prepared for a human to approve.
Core infrastructure hosted in the EU · answers with source citations · no ad or analytics trackers
Ada is an AI assistant. You are talking to a system, not a person.
On your website
Six things Ada does for your customers
This works today, on your website and your content. Each item has its own page with a sample conversation and a clear split between what is ready and what is still being built.
She books and runs meetings
A slot picked inside the chat window, a confirmation with a calendar file, reminders. She runs the demo herself, prepared on the visitor's own website.
She keeps the e-mail thread
She receives and replies in the same thread, from the same knowledge as on the website, and protects booked slots with reminders.
She surfaces the right material
From a catalogue you define, at the moment it answers the question just asked. She never invents an address of her own.
She records what customers looked for
Questions without coverage become a list of topics to write, and every turn leaves a trace in the audit log.
She learns how you work
You show her once on a recording how a repetitive task is done, and Ada takes it from there. Before any step that cannot be undone, she waits for your click.
Where this behaviour comes from: the platform and Ada's working cycle.
Ada learns
Not every job can be written down as a procedure. Some of the work in a company lives in the heads of the people who do it, and the simplest way to explain it is to show it on screen. That is how Ada learns it.
You show her once
You record your screen doing the task the way you always do. No narration, no slowing down. The recording stays on your computer, and only the frames you approve reach Ada.
Ada breaks it into steps
You get a list of steps in plain language, marking which of them change data and which cannot be undone. You fix what needs fixing and publish.
From then on she does it herself
The process starts on your click, at a set time, or when a message arrives that triggers it. After every run you get a report of what was done.
How it works step by step and where Ada's independence ends: the Automations module.
Inside your company
Connect your stack. Tell Ada what you need. Ada operates.
No new windows and no new passwords. Ada comes to the tools your team already uses.
Connect your stack
You add Ada to your Slack or Teams workspace like a new team member and point her at the knowledge groups she may access. No changes to your infrastructure.
Tell Ada what you need
You write to Ada in a channel, in a thread or in a direct message, the way you would write to a colleague. The answer comes back in the same place, with the knowledge base sources cited.
Ada operates, you approve
Ada does not send you off to read the documentation. She comes back with a finished answer or an artifact, and every change to your data waits for your click on a confirmation card.
Use cases
One AI employee. Every function.
Ada is not a chatbot for a single team. The same machinery, a knowledge base with citations and confirmation cards, serves support, finance, sales and engineering.
Support and onboarding
A new hire asks Ada in a direct message about benefits, leave and procedures. Ada answers from your company's curated knowledge base and points to the source every time. Questions with no coverage land in the knowledge gap register, so HR sees what is still missing. The result: a finished answer in the thread with a link to the source, instead of a pointer to the intranet.
Finance
The team asks Ada in the finance channel about the expense policy, settlement deadlines and purchasing procedures, and she answers from approved documents, with citations. Questions with no coverage land in the gap register, so accounting sees which procedure is still unwritten. The result: fewer questions for accounting, and answers that follow the current procedure rather than somebody's memory of it.
Sales
The team asks Ada about pricing, terms and proposal wording, and she answers only from approved material, with citations. The rep gets the current wording of a clause in the thread, not the version from an attachment sent a quarter ago. The result: customer answers consistent with your offer, and one price list instead of ten copies.
Engineering
Questions about on-call procedures, runbooks and decisions from past reviews get an answer sourced from the knowledge base, in the channel or in the incident thread. A question with no coverage lands in the gap register, so the team sees which runbook is still unwritten. The result: the on-call engineer gets the procedure in the thread instead of searching the wiki at three in the morning.
In progress
1,600+ tools through managed connectors
We are building the connector layer through which Ada will connect to more than 1,600 tools: CRM, project management, finance and accounting, communication, analytics, customer support. Today Ada works in Slack and Teams on your knowledge base. Below is what arrives with the connectors and recurring tasks.
She will pull revenue from Stripe, check the sales pipeline in Pipedrive, review open issues in Linear and put it all together into one picture. In a single Slack message.
One-click OAuth
Most connectors will take one click to authorise: you pick the tool, confirm access and you are done. You set the permission scope when you connect, and you can disconnect at any moment.
Notify meAPI keys where they are needed
Some tools use API keys instead of OAuth. Either way Ada will handle the authentication, and secrets will stay on the platform side, never in the conversation.
Notify meNo webhooks and no zaps
Connectors will be managed on our side, with no webhooks to configure and no Zapier flows, and every write action in your tools will go through a confirmation card.
Notify meRecurring tasks
The same instruction will be schedulable as a recurring task that comes back at a set time: a weekly summary for the team or a reminder before a deadline. Today Ada acts on instructions given directly.
Notify meFinance: a summary from your payment tools
Each month Ada will pull data from your payment and accounting tools, turn it into a summary and post it to the finance channel as a recurring task. Sending an invoice or correcting a record will go through a confirmation card.
Notify meSales: the opportunity update in your CRM
After a sales call Ada will prepare the opportunity update in your CRM and put it up for one-click approval. A CRM that stays current without manual clicking.
Notify meEngineering: morning triage
Ada will collect overnight monitoring alerts and tracker issues, group recurring problems and post a morning summary to the team channel. One thread instead of thirty notifications.
Notify meWhen the topic outgrows chat, switch to voice
Not every matter is settled in writing. Ada sends a link to a conversation room and speaks in the same voice she writes in.
A conversation room with an avatar
From the chat you get a link to a room where Ada speaks and listens in the browser, with a live avatar. No installation, no new account and no new passwords. One click starts the call.
The same rules as in chat
Voice is not a separate mode with looser rules. The same knowledge sources, the same limits and the same permissions of the person asking apply exactly as they do in writing.
Ada joins your Teams meeting
You do not have to move the client into our room. Ada joins the meeting you already have scheduled, with a face and a voice. The face and the speech are produced by us, while entry to the meeting itself goes through a Microsoft service.
A participant, not a recorder
Ada has her own tile, listens to the conversation and answers out loud. She enters the way any guest from outside your organisation does: she appears in the participant list under her own name, and if you keep the lobby on, she waits there to be admitted. No recordings are made, because our platform has nothing to record with.
Silent until you call her
In a meeting of several people, an agent that answers every sentence is unbearable. Ada listens by default and speaks when someone addresses her by name. Talking about prices next to her is not an instruction to her.
She runs the meeting, she does not just sit in it
When Ada is the one running it, she does not wait to be called. She opens the meeting, sets out the plan, moderates the conversation and hands the floor back when her part is done. She shares her screen and paces the presentation with what she is saying, the way an employee would.
She sees what you are showing
She looks at the shared screen and refers to what is on it. She knows the participant list, so she addresses people by name instead of speaking into the room.
A presentation you can interrupt with a question
Ada shows your deck and talks through it slide by slide. At the end she stops the show and asks for questions. When someone asks about an earlier slide, she goes back to it, puts it on screen and answers against it, then returns to where she left off.
After the meeting, Ada does not disappear
What you agreed does not have to wait for someone on the team to find a free evening. Ada has her own account in your company and keeps working once the conversation ends.
Her own account, and only the resources you grant
Ada gets an account in your company directory, like any new employee. You grant her permissions on your side and withdraw them in the same place. You point her at specific SharePoint libraries, and she reaches for nothing beyond what you share with her.
Company knowledge instead of guesswork
Your documents become her knowledge base: procedures, price lists, service descriptions, notes from rollouts. She names the document she is drawing on, so every sentence she says can be checked.
Tasks out of what was agreed
When a decision is made in the meeting, Ada records it as a task with a deadline and an owner. Only the people you designate may assign her work, and only within the scope you set for them.
She comes back with a report and a proposed time
She writes to the people responsible for a topic herself, collects their answers and comes back with a summary. If the matter needs a conversation, she proposes a time from the free slots in the calendar and sends the invitation.
Security
The security behind your AI employee
An AI employee gets exactly the same boundaries as a human one. These are not settings you can overlook, they are the way the platform works.
What Ada does
- Answers only from your company's curated knowledge base, citing sources
- Records questions with no coverage in the sources as knowledge gaps
- Issues a confirmation card for every action that changes data
- Keeps an audit trail without conversation content
- Runs its core infrastructure in the European Union, on servers in Warsaw
What Ada does not do
- Does not train models on your data
- Does not change anything without human approval
- Does not answer from the internet or from guesswork
- Does not show content to people without permission
- Does not mix data between companies
Gates before the model, not after it
Every request passes availability, mode, limit and knowledge corpus checks in turn before it reaches the language model at all. When any gate blocks, the model provider is never called.
Citations or rejection
The answer is buffered in full and checked for coverage in the sources before anyone sees the first character. Missing citations mean the whole answer is rejected, and the question goes to the knowledge gap register.
A human approves every change
Data changes never leave the model's turn. Ada prepares a confirmation card, a human approves it with a separate click, and the safety checks run again at the moment of approval.
The asker's permissions
Ada answers only from content the person asking is allowed to see. This is not a filter applied to the result, it is a condition of the query itself. Separation between companies is required on every database query and covered by isolation tests.
Failure closes, it does not open
Budgets and limits are fail-closed: when something breaks, the feature shuts down rather than running unchecked. On top of that, rate limits for voice conversations and SSRF protection when importing content from URLs.
EU infrastructure, no training
Core infrastructure runs on servers in Warsaw, and on the Enterprise plan we sign a data processing agreement (DPA). Models run through the commercial APIs of the providers listed in our subprocessor register, under no-training terms; some of that processing may take place outside the EEA under appropriate safeguards. The site and the widget use no cookies, no ad or analytics trackers and no external CDN for our code.
You do not have to take our word for it. The platform
publishes a capability matrix at /capabilities, where you can
check for yourself what is enabled on your instance.
Questions about Ada
How does Ada know what to answer?
Only from your company's curated knowledge
base. Every answer cites its sources, and the coverage check runs
before the answer is shown, not after the fact. Ada does not reach
out to the internet and does not fill gaps with guesswork.
What happens when Ada does not know the answer?
She says plainly that she does not know, and
the question goes to the knowledge gap register. You see exactly
what is missing and decide what to add. That register is one of the
most practical sources of insight into what your team is actually
looking for.
Can Ada change something in our systems on her own?
No. Every write action goes through a
confirmation card approved by a human. A change never executes from
the model's turn, and the safety checks run again at the moment of
approval, not only when the card is prepared.
Who sees which answers?
Ada answers only from content permitted for
the person asking. Audience groups set the scope at the level of
the knowledge query itself, so content outside those permissions
never enters the answer. Ada does not hide it, she simply never
sees it.
Is our data used to train AI models?
No. Models run through commercial APIs under
no-training terms, and your content stays yours. Conversations,
documents and Ada's output feed no training process.
Where is our data processed?
Core infrastructure, the database and voice transport run in the EU, on servers in Warsaw. AI features rely on the providers listed in our subprocessor register and some processing may take place outside the EEA under appropriate safeguards.
On the Enterprise plan we sign a data processing agreement (DPA)
and set a conversation retention period that matches your
policy.
How does Ada connect to our tools?
Today through Slack and Teams, where she
works like a team member, and through your knowledge base, which
she answers from with citations. Links to external tools such as
CRM, finance or project management are being built: their state
is described in the block marked In progress on this page and on
the roadmap. Every write action in your tools will go through a
confirmation card.
How do we check what our instance can actually do?
The platform publishes a capability matrix at
/capabilities. It is a technical inventory of your
instance's features that you can verify yourself, without taking our
word for it.Hire Ada onto your team
Tell us what tools your team uses and what it works on day to day, and we will show Ada in action on your own scenarios. No pressure and no commitment.