Finance
The same question, answered three ways
Each person asks the chat their own way and gets a generic answer, without the company's credit policy, without the order system and without knowing what needs approval.
One agent for the area
Platform · Agents and teams
Whoever knows the routine builds the agent. The team uses it.
An agent is an assistant with a name, instructions, documents it always consults and the tools the company enabled for it. Whoever knows the routine builds an agent for the area; a team gathers several agents under a working mode; and what was built stays available to the right people, with a defined role.
See how it worksAUTOMATIC · ADVANCED · PREMIUM
the AI level is the agent's choice; Skyller picks the model
PERSONAL · COMPANY · PLATFORM
your agents, your company's and Skyller's ready-made ones, on the same screen
TOOL BY TOOL
each agent only uses what was switched on for it, asking for approval whenever the company wants
4 TEAM MODES
coordination, routing, broadcast and tasks: several agents, one goal
Where this shows up in your day
The knowledge of the routine lives in one person's head.
Every area has someone who knows how the work has to be done: which document counts, which system to check, what cannot be promised. When that person is away, the answer becomes improvisation — and generic AI improvises along.
HR
Explaining the policy for the tenth time
Onboarding, leave, remote work: the questions repeat and the answer depends on who was asked. The approved document exists, but nobody checks it before answering.
Answers by the policy
Operations
Work that passes through several hands
Gathering data, checking against the procedure, writing the summary and flagging open items are steps with different specialties. One do-it-all assistant does each step halfway.
One team, several steps
How it works
From what one person knows to an agent the team uses.
Five steps, no programming. Whoever builds it chooses what the agent knows, what it may do and who may use it — and can start by describing the routine in one sentence.
01
Describe or build
Create with AI: you describe the role and Skyller writes the name, persona, instructions and opening messages for you to review. Or build it by hand, in five steps.
02
Give it knowledge
Choose the documents the agent always consults and the hub where it works. Whatever is published and valid gets in; drafts do not.
03
Switch on the tools
Among the systems the company connected, switch on only what that agent needs — and mark what must ask for approval before acting.
04
Choose the AI level
Automatic, Advanced or Premium. The agent carries the level; Skyller picks the model, from providers such as OpenAI and Anthropic, for each task.
05
Share or publish
Invite people or groups as viewers or editors. An administrator can make it a company agent, visible to the whole organization.
Live proof
One request, three agents, one answer with its source.
A team is not a bigger chat: it is a leader splitting the work among specialists and assembling the result — and whoever asked sees each part happening.
Starting point
A “Monthly close” team in Coordination mode, with three agents: one consults the Finance hub documents, one checks the reconciliation procedure, one writes the summary.
What happens
- 01In the Finance hub chat, the person picks the team as the target and asks: “Prepare the summary of September's open items.”
- 02The team leader splits the request: one member finds the published procedure and cites the source, another lists the open items, and the conversation panel shows each member working.
- 03The leader merges the answers into a single summary; where the procedure was used, the document name appears as a link.
Result
An answer assembled by different specialists, in the same chat — and whoever asked sees who did what, instead of receiving a generic text.
What this scene does not promise
- The team works with what each member can access: documents published in the hub and enabled tools. Without them, it answers with what it has and says what was missing.
- An action in a connected system still asks for approval when classified as sensitive — a team gets no extra autonomy for being a team.
- A team is coordination of agents, not an autonomous digital employee: every run starts from a request or from an automation with an owner.
To evaluate calmly
What an agent carries, and who is in charge of it.
Select a topic to see its capabilities, conditions, and limits.
- Instructions, persona and opening messages
- The agent has a name, avatar, description, category, instructions and conversation suggestions. All editable later — including what Skyller wrote during AI creation.
- Fixed knowledge base
- Documents the agent always consults, chosen among the published ones, plus what it finds in the conversation's hub and workspace. Drafts and expired documents stay out.
- Tools switched on one by one
- Each agent has its own tool set within what the company connected, with an on/off per item and an approval policy: inherit, always ask or never ask.
- AI level and reasoning
- Automatic lets Skyller pick the model; Advanced and Premium fix the range. The reasoning level follows the mode. The model itself comes from providers such as OpenAI and Anthropic.
- Four scopes on one screen
- Personal, shared with you, the company's and Skyller's ready-made ones appear together, with filters by scope and category. The company's default agent answers when nobody picks another.
- Roles in sharing
- A viewer uses; an editor adjusts instructions and tools; only the owner changes the scope. An administrator decides what becomes a company agent; the agent manager looks after their own without giving away the organization's assets.
- Duplicate to adapt
- Duplicate an agent to adapt it your way without touching the original others use. The company default stays whatever the administrator set.
- Teams with four modes
- Coordination: the leader splits, delegates and synthesizes. Routing: picks one member and delivers that member's answer. Broadcast: everyone gets the same task. Tasks: the leader decomposes, assigns dependencies and reviews.
- Teams inside teams
- A member can be another team. The leader has its own AI level and, if the company wants, its own tools; without them, it only delegates.
Real application
The finance agent, built by the finance person.
The gain is not one more robot: it is the right way of doing the routine ceasing to depend on one person and becoming a resource of the area, with visible rules.
- 01Finance analyst
Describes it in one sentence: “agent that answers about the credit and collections policy and checks orders in the system”. Skyller proposes a name, persona and instructions; she reviews and saves.
- 02Finance analyst
Picks the published Credit and Collections Policy as the fixed base and places the agent in the Finance hub.
- 03Administrator
In the already connected order system, switches on only “look up order” and “record payment”; marks the payment as “always ask for approval”.
- 04Finance analyst
Shares with the Finance group as viewers and with the coordinator as editor.
- 05Finance team
Asks in the hub chat. Look-ups come back right away; recording a payment opens the approval card before it happens.
Comes ready
- Creation with AI or by hand, full editing afterwards
- Tool scope per agent with an approval policy per item
- Four team modes and a panel of members working in the conversation
- Sharing with roles and Skyller's ready-made agents by area
The company decides
- Which agents become company assets and which one is the default
- Which systems are connected and which functions each agent may use
- Which actions ask for approval: per company, per agent or per conversation
- How many agents each person may create, according to the plan
Questions from whoever decides
What people ask before building the first agent.
Next step
Build the agent for the routine that drains your area most.
Describe the role in one sentence, pick the document that counts and switch on only the necessary functions. Share it with two people and watch the answer come out the same for both. Or schedule a demo and build the first one with our team.