Customer service
“Do we charge for this or not?”
The rule exists, it lives in an approved procedure and nobody remembers which one. Asking in the group chat costs two people's time and still comes back from memory.
Answer with its source
Platform · Chat
The models you know. The context only your company has.
Skyller uses AI models from providers such as OpenAI and Anthropic to write the answer. What she adds is everything around it: the place where the conversation happens, the documents your company approved, the systems it connected, the access of whoever asked, and confirmation for actions that require approval.
See how it worksAUTOMATIC · ADVANCED · PREMIUM
you pick the level of the task; Skyller picks the model
UP TO 5 FILES PER MESSAGE
PDF, Word, PowerPoint, Excel, text, image and audio
A SOURCE THAT OPENS
if a document was used, its name becomes a link to the document
ASKS BEFORE ACTING
an action that touches a system waits for you to approve or deny
Where this shows up in your day
The question that today becomes a message to a colleague.
This is not generic AI answering anything about everything. It is the specific question of your work, answered with what your company wrote, approved and connected.
Administration
The contract arrived as a PDF
Thirty pages to find the term, the penalty and the adjustment. Pasting it into a personal chat takes the document out of the company's control and does not solve the next time.
Summary with the quoted passage
System routine
Check it without opening another system
The information is in the ERP, the CRM or the system only your company has. Whoever needs it does not always have a login — or loses the morning looking for where to click.
A check through the conversation
How it works
Five choices before the answer.
None of them require knowing what a language model is. They are work choices: where, with whom, with which material and with how much autonomy.
01
Where
The conversation is born inside a hub and a workspace. What applies there — instructions, documents, team — starts applying to the conversation.
02
Who answers
An agent for the area, chosen in the conversation itself. A team of agents in coordinated steps or a flow already built come in through that team's or flow's link, and the conversation then shows who is doing the work.
03
What you bring
Text, attached files, a screen capture and the tools you enable for that conversation in the tools panel.
04
Where it came from
When the answer uses a document from the knowledge base, the file name appears as a link and opens the document — no code or identifier in front of you.
05
How far she goes
Delete, remove, archive, send: before executing, Skyller shows a card with what she is about to do and waits for Approve or Deny. Editing and registering do not stop by default — and your company can require the brake on those too, function by function.
Live proof
Ask for an action. She asks for your confirmation.
This is the difference nobody sees in an ordinary chat: Skyller does not only answer, she acts in the connected systems — and she does not act alone when the action changes something.
You ask
“Cancel order 4471 in the system and let the customer know.”
What happens
- 01Skyller identifies the tool of the connected system and assembles the call with the data from the conversation.
- 02Instead of executing, she opens a card in the chat with the action and the parameters, and stops there.
- 03You approve or deny. Only after Approve does the action go out, and the result comes back into the conversation.
Result
The action exists, the decision is yours, and both are recorded in the conversation — instead of a “confirm?” in plain text that nobody can audit later.
What this scene does not promise
- Skyller only acts in a system the company connected and enabled function by function; without a connection she reads and writes, she does not execute.
- The conversation sees what the person asking can see — the same question may have different answers for different people.
- Not every action opens a card: the confirmation applies to the ones classified as sensitive, not to a read-only lookup.
To evaluate calmly
What chat does, with the limits attached.
Select a topic to see its capabilities, conditions, and limits.
- Three levels, not three brands
- Automatic for everyday work, Advanced for heavier reasoning and Premium for what is critical. You pick the level; Skyller picks the model. The level applies to the whole conversation: to change it, start a new one.
- Models from the people who make models
- Behind the levels are models from providers such as OpenAI, Anthropic, xAI, DeepSeek and Z.AI. Skyller builds none of them and does not compete with them on general intelligence: what she does is pick which one to use for each task and hand it your company's context.
- Tools per conversation
- A panel shows the tools and skills available in that session and lets you add or remove them without touching the agent. What the company has not enabled does not appear.
- Teams and flows on the same screen
- Besides a single agent, the conversation can target a team — several agents in coordinated steps — or a flow already built. The screen is the same; what changes is who does the work.
- Files with a clear rule
- Up to five files per message: documents up to 50 MB, images up to 5 MB and audio up to 20 MB, adding up to 100 MB. PDF, Word, PowerPoint, Excel, HTML, text, Markdown, CSV, images and audio. Video is not accepted yet.
- An attachment is not knowledge
- The file you attach applies to that conversation. To become the official answer for the whole team it has to enter the company's knowledge base, with an owner, a version and approval — which is another door.
- A source when there is one
- When the answer uses a published document, the file name comes as a link and opens the source. When there is no document behind it, Skyller answers without inventing a citation: if a specific fact about your company is missing, she says it is not available instead of filling the gap with general knowledge.
- History that continues
- Conversations are saved, with attachments and decisions. You pick up where you left off and follow the credit usage of that conversation.
- Sharing means reading
- A conversation can be shared with colleagues for reading — up to one hundred people per resource. Whoever comes in this way follows the history, but does not approve actions in your place.
Real application
A contract question, from start to finish.
The same request that today goes through three people and two days, solved in one conversation — with a record of who decided what.
- 01Administrative assistant
Opens the contract workspace, attaches the PDF she received and asks what changed compared with the previous contract.
- 02Skyller
Compares it with the published contract policy, answers the differences and cites the policy as a link that opens the document.
- 03Administrative assistant
Asks her to record the legal review as a pending task on the team board.
- 04Skyller
Shows the card with the task she is about to create and waits. After Approve, she creates it and returns the card link.
- 05Coordinator
Receives the shared conversation, reads the history with the source and the decision, and carries on from there.
Comes ready
- Choice of AI level, attachments, history and conversation inside a hub and a workspace
- The document cited as a link when the answer uses published knowledge
- An approval card before sensitive actions, recorded in the conversation
- The conversation's tools panel and read-only sharing
The company decides
- Which systems stay connected and which functions of each one the AI may use
- Who may use the Premium level and what each person's credit limit is
- Which documents are published and valid — only those enter the answer
- Which agents, teams and flows exist and who has access to each one
Questions from whoever decides
What usually holds the decision back.
Next step
Start with a question you repeat every week.
Open a conversation in your area's space, attach the document you consult the most and ask the usual question. Or book a demo: the team builds the scene with your documents and fictional data.