Sign in Start free trial
Platform

 AI

An assistant that prepares work in every hub; a person always decides.

 AI is the assistant built into every hub. It answers questions, drafts records, configures screens from plain language and plans multi-step tasks, always as the signed-in person and within their rights. Money, approvals and other sensitive decisions stay with people, and administrators control what AI may do, which models it uses and how much it can spend.

45 features available 6 planned 7 capability areas
Highlights

What stands out

Draft, never decide

The assistant prepares; people approve, reject and release.

Say it, and the screen changes

Hide a field or add a column in plain words with one-step undo.

Stop everything in a moment

Kill switches halt one agent or all AI for a tenant.

Spend you can predict

Allowance, alerts and prepaid credits keep AI costs bounded.

Your rules for AI

Autonomy, fields, providers and regions are set by you.

Capabilities

Every feature, in one list

Search or filter. Planned items are labelled; everything else is available now.

Assistant in every hub

Ask, find, draft and act from the screen you are on.

12 features
  • Context-aware chatThe assistant knows the hub and record you are in, and acts as you with your permissions.
  • One action set for chat and external toolsThe same actions power the chat and the MCP gateway, so behaviour is consistent.
  • Plain replies with action chipsConfirm, update existing, cancel, take me there and undo chips; no raw codes or dates.
  • Follow-upsSay mark it paid and it binds to the record just shown.
  • Fast path for common commandsSlash commands and deterministic matches answer without calling a model.
  • Pre-write awarenessWarns about duplicates and conflicts before saving.
  • Grounded numbersFigures come from governed derivations, not model arithmetic, with a view-report link.
  • Untrusted content kept apartDocuments and e-mails are treated as data, never as instructions.
  • Every UI action has an AI actionA guard checks that every write screen has an assistant counterpart or a declared reason.
  • Saved conversation history and searchPlannedBrowse past threads with retention set by tenant policy.
  • Voice input and spoken repliesPlannedSpeech in and out with confirmation still on screen.
  • Multilingual chatPlannedReplies beyond English; English only until release.

Configure by plain language

Tell  AI how you want a screen or a setup to look.

7 features
  • Personal view changesSay hide this field or show that column; changes apply to your own view with one-step undo.
  • Company changes with previewPropose a change for everyone, see a preview, confirm; required fields are never hidden.
  • Feedback with screen contextSend feedback from any screen with the context attached.
  • Configuration through your own AI toolsCustomers and implementers can configure through the admin MCP and the command-line tool using their own key.
  • Plain-language drafts in buildersDescribe a workflow, automation, process or pipeline and get a draft for a person to check.
  • Intelligent completionSuggested values in forms, grids, command prose and data mappings that a person accepts.
  • Auto-fill from contextLocale and context populate fields so people type less.

Multi-step tasks and agents

Ask for several things at once; review the plan; confirm.

7 features
  • Action plansA request becomes a typed plan of existing actions, previewed with every value.
  • One confirmation for safe batchesMoney, destructive and approval steps still need their own confirmation.
  • Step-level safety and undoEach step runs through licence, permission, policy and audit; a failure is reported clearly and undo works where a step has an inverse.
  • Cheap model first, escalate when neededA low-cost model handles routine requests and escalation to a stronger model is metered.
  • Tenant-defined agentsGive an agent a goal, a trigger, allowed actions and guardrails; it runs as its own identity and leaves drafts for people.
  • Contract agentsAgents that draft intake requests, renewal briefs and reminders, off by default, with their own budget and halt.
  • Autonomous action within policyPlannedLow-risk actions without a person in the loop, inside limits.

AI policy and governance

Administrators decide what AI may do, per hub, role and user.

10 features
  • Autonomy levelsSet off, read only, draft, act with confirmation or act within policy by hub, sub-module, entity and action.
  • Policy by role, group and userFor example, block AI from salary data for most roles.
  • Field exclusionsFields never sent to a model.
  • LimitsCaps on amounts, record counts and bulk size.
  • Provider and region allow-listRestrict which providers and regions are used, or turn AI off.
  • Single-use confirmation tokensA sensitive action needs a confirmation bound to the exact request.
  • Policy decision logEvery decision is recorded append-only.
  • Same policy for external AI toolsMCP clients are governed exactly like the in-app assistant.
  • Bring your own model keyStore your own provider key, sealed, with fresh step-up and an audit trail.
  • Policy simulatorPlannedAsk what the sales assistant is allowed to do.

AI control tower

Treat every non-human caller as a registered agent you can stop.

7 features
  • Agent registryEvery agent, key and client is registered with a manifest; unregistered callers are refused.
  • Risk scoring and tiered approvalAgents are scored and approved by tier through workflow.
  • Kill switchesStop one agent or all AI for a tenant, including in-flight calls, with fail-closed behaviour.
  • Resume under supervisionResuming needs the owner with fresh step-up and starts a probation.
  • Activity feed and timelineMetadata-only events with cost per agent and CSV export.
  • Never listActs AI may never perform are listed, enforced and proven by tests.
  • Person-only decisionsApprove, reject, publish, release, send to customers and money movement are not available to AI.

Cost controls

Predictable spend with no surprise bills.

6 features
  • Allowance and fair-use poolAI usage is metered through one choke point, with an allowance and thresholds.
  • Hard stop by defaultWhen the allowance is used up the assistant pauses and says so, with optional top-ups.
  • Monthly budget alertsSpend telemetry with alerts against a monthly budget.
  • Cost per agentSee what each agent costs.
  • Prepaid credit walletCredits are drawn in order of expiry from an append-only ledger; AI cannot buy credits.
  • Usage dashboardUsage per tenant and period.

Builder recognition

Recognise people who configure  AI well.

2 features
  • Builder recognition engineAn append-only record of configuration actions with badge and level definitions.
  • Badges and levels screensPlannedVisible badges and levels for builders.
Assistant

What  AI prepares here

It prepares the work. A person confirms it.

  • Answer a question about a record or a report
  • Draft a record, quote, contract request or message
  • Plan several actions and run them after one confirmation
  • Hide, show or reorder fields on a screen for a person or the company
  • Draft a workflow, automation, process or pipeline
  • Suggest values while a person fills a form or a grid
  • Prepare a configuration change for review
  • Create an agent from a goal for a person to enable
  • Show how much AI allowance is left
  • Explain a decision or a failed delivery
Inside every hub

Workflow, reporting, integration, migration

The same four building blocks, specific to boss AI.

Workflow

AI can draft policies and automations but cannot approve, reject, delegate or publish; agent approvals by risk tier go through workflow.

  • A new agent above a risk tier needs approval before it runs
  • AI drafts a policy; a person checks and publishes
  • Company-wide screen changes need a confirm

Reporting and analytics

AI usage, spend and agent activity are available as governed datasets.

Datasets

AI usageAgent activityAI policy decisions

Dashboards

AI usage and cost

Integration

External AI tools connect over MCP and the admin MCP under the same policy.

/v1 AI usage and agent resources; /mcp and /mcp/admin

Connectors

MCP

Migration

Migration includes an AI pack drafter that proposes connector packs for a person to approve.

Mobile

On the phone

The phone frame has an AI tab; the same assistant answers there. Voice is planned.

Who uses it

Every userAdministratorCompliance and securityFinance owner of AI budgetImplementerDeveloper using an external AI tool
Questions

Frequently asked

Can AI approve or pay?

No. Approvals, rejections, publishing, releasing and money movement are person-only.

Whose rights does AI use?

The signed-in person's. It never has more access than they do.

Can we turn AI off?

Yes. AI can be off for the tenant, a hub, a role or a field, and there is a kill switch.

Which model providers are used?

Administrators can restrict providers and regions, or add their own key.

What if we use up the AI allowance?

By default the assistant pauses and tells people; top-ups can be enabled.

Can people customise screens with AI?

Yes, personally with undo, or for the company with a preview and confirmation.

Does AI use our data to train models?

The design aims for no training on tenant data; confirm with your agreement.

See  AI on  AI

Talk to us