Travis — the AI operator Live
Embedded AI assistant that answers questions and executes business actions through a permissioned tool registry, with approvals, tenant isolation, autonomy modes, and cost-aware model routing.
What Travis can do
- Conversational chat with business context
- Read tools (customers, jobs, invoices, leads, equipment, employees, revenue/analytics summaries)
- Create/update via permissioned write tools
- Communication tools (email/SMS/contract/invoice/review) — always require approval
- Routing to EstimateAI, Landscape Design, InstaSign, and pages
- Cost-aware model routing across tiers
What Travis cannot do (limitations)
- Always requires human approval for money movement and all customer-facing sends (NEVER_AUTO_APPROVE set) — cannot be bypassed even in full-auto mode
- Free = read/basic only; Starter = no Pro+ actions; tenant-scoped to one business
- Per-business rate limits (10/min, 60/hour) and monthly AI token budgets (Free 0 / Starter 100 / Pro 500 / Enterprise 2000)
- Max 5 tool-calling rounds per turn; 50% monthly-token burn cap per session; per-customer 2 automated messages/day
- Blocked by missing integrations (Gmail/Twilio/Stripe) and by abusive-content filter
- No payroll disbursement, tax calculation, or live website publishing without entitlement
Approvals and safety
The following actions always require human approval and cannot be auto-executed in any mode:
- sendInvoice
- sendContract
- sendSms
- sendEmail
- chargeCard
- issueRefund
- markInvoicePaid
- createJob
- autoBookJob
- and the full NEVER_AUTO_APPROVE set
Model routing
Travis routes each request to the most cost-effective capable model:
| Default chat | gpt-5.4-mini |
|---|---|
| Complex / multi-tool | gpt-5.4 |
| Escalation / recovery | gpt-5.5 (at most once per request) |
| Routing / classification only | gpt-5.4-nano (classification only; never conversational) |
See automation modes and Website Leads.