A different model for every department.
Put Finance on deep reasoning, Comms on something fast and cheap, and Support on a model running inside your own building. Eight providers, twenty-six models, eleven departments, changed from one screen and live on the next prompt.
The assistant opens with your pipeline beside it: active deals, and the ones going quiet, named, with how long. A blank box waiting for a prompt would know none of that.
Routing
Every department, its own model.
Most AI in business software gives every team the same fixed model from one vendor. Here each department gets the model that suits its work, and you can change it whenever you like.
- SalesGPT-5.4Fast, and it talks to people all day
- FinanceClaude Opus 4.8Reasoning, where being wrong costs
- MarketingClaude Sonnet 4.6Writing
- OperationsGPT-5.4Builds your workflows
- EngineeringClaude Opus 4.8Long context
- CommsGemini 3.5 FlashHigh volume, every message
- SupportLlama, on your own machinesNothing leaves the building
- PeopleGemini 3.5 FlashEveryday questions
- WebGemini 3.5 FlashContent and tidy-ups
- AnalyticsGPT-5.4Turns a sentence into a chart
- NearSync AIMasterInherited by every HQ on Master
Change one and it applies to every session on the next prompt. A department left on Master follows whatever the base model is.
Change it once, and it applies on the next prompt.
Your models
Eight providers, or your own hardware.
OpenAI, Anthropic, Google, xAI, Mistral, Groq and DeepSeek, twenty-six models between them. Or run Llama on machines you own, so nothing leaves your building at all. Set one as the base model and every department uses it until you say otherwise.
What every department uses until you give it one of its own. Change it once and they all update together.
Spoken conversations run on a realtime model chosen separately, so talking is fast without making everything else expensive.
Point a department at Llama on your own hardware. Nothing about that department's work leaves your network.
Ask two models the same thing, side by side.
Watch them answer in parallel, see how long each took and what each cost in credits, and keep the one you preferred.
- Two or three at once. Pick the models you are weighing up and send one prompt to all of them.
- They answer side by side. Each answer appears in its own column as it is written, so you can compare as you go.
- The same question, exactly. Every model gets an identical prompt, including the department context, so the comparison is fair.
- How long each took. Real elapsed time per model, measured, not estimated.
- What each one costs you. Shown in credits, so you can see whether the better answer is worth the difference.
- Pick the winner. Choose the answer you preferred. The comparison is saved, so you can look back at it later.
Name your assistant and shape how it speaks.
Set it up once and it looks and sounds the same everywhere it appears: in chat, the inbox, agents and voice.
- Give it a name and pick from fourteen icons
- Write its role and the greeting it opens with
- Set its tone, how long it answers for, and its language
- Add your own rules in your own words
- Name topics it should stay off, and ask it to cite its sources
Identity
Preview it as you build it.
Fill in the name, the icon, the role and the greeting on the left, and a live preview on the right shows exactly what your team will see, updating as you type. There is nothing to publish before you can check it.
Persona and safety
Ten things you set once.
These shape every answer the AI gives anyone in your company, on every screen it appears on.
Call it whatever suits you. The name follows it everywhere, in chat, in the inbox, on agents and in voice.
Fourteen marks to pick from, so it looks like part of your product rather than a bolted-on robot.
One line describing what it is there for. It appears under the name on every screen.
The first thing it says when somebody opens a new conversation.
Professional, warm, direct, or whatever you write. It follows this in every reply.
Short and to the point, balanced, or thorough.
Ten to choose from: English, Arabic, French, Spanish, German, Portuguese, Hindi, Chinese, Japanese and Korean.
Anything else you want it to always do, in your words.
List the subjects it should not discuss with your people.
Ask it to cite the knowledge base article behind an answer, every time.
Knowledge
Teach it what your company knows.
The AI answers from your own knowledge base articles, finding the right one by meaning rather than by exact keyword. An article only reaches customers after somebody publishes it and marks it customer facing, and both of those are off until a person turns them on. You can refresh the whole library whenever you like.
An answer names the article it came from, so anyone can check it.
When it is not confident in an answer, it says so and offers a person instead of guessing.
Two separate switches, both off by default, before an article can reach a customer.
Keep the prompts that work.
A shared library your whole organisation can see, organised by department, with ten ready to use the day you start.
- Deal risk, follow-up emails, expense categories, collection letters and job descriptions, ready on day one
- Plus performance reviews, content calendars, site audits, written procedures and meeting summaries
- Write your own, with placeholders you fill in each time
- Your prompts work exactly like the built-in ones, in the same list
- Filter by department so everyone only sees the prompts they need
Access
Who can do what.
Everybody uses the AI. Choosing which model a department runs on, and how the assistant speaks to the whole company, belongs to Admin.
See which model your department runs on.
Use the assistant, and the prompts in the library.
Add prompts to the library for your team.
Choose providers, route departments, and set the persona.
Admin is the highest role and is not constrained by the permission grid, so it is worth keeping to the smallest group that genuinely needs it.