Miles isn't a chatbot — it's a team of specialists that qualifies leads, matches inventory, books appointments, scores every opportunity, and answers your reviews. Autonomous, 24/7, across text, web chat, and email.
Most tools in the dealership space bolt a chat window onto a help doc. Miles operates the way your best BDC rep does — it carries the goal from hello to a booked appointment.
Each agent owns one job and does it well. Miles' orchestrator routes every conversation to the right specialist — and the right Claude model for the task.
Reads every inbound message and decides which specialist should own the reply. Emits structured routing only — it never calls a tool itself, so the hand-off stays fast and auditable.
Works budget, timeline, trade-in, and buying intent into the conversation the way a good BDC rep would — then passes a qualified lead down the line.
get_lead_context · qualify_leadSemantic search across your live inventory with pgvector. Matches what the customer actually described — body style, budget, use case — not just exact keywords.
recommend_inventoryTurns interest into a booked appointment with conflict-aware scheduling, then syncs it straight back to your CRM so the showroom is ready.
book_appointment · qualify_leadScores every lead 0–100 the moment it lands — engagement, intent, responsiveness, recency, and source quality — with a close probability and a predicted timeline.
Drafts on-brand replies to Google, Yelp, and Facebook reviews. Reads sentiment, picks a tone, and leaves the publish button to your team.
The customer talks to Miles. Behind the single thread, the orchestrator hands the turn to whichever specialist fits — qualify, match, then book — without the customer ever feeling a hand-off.
Semantic matching across your live lot. A customer asks for "something safe for the family that won't kill me at the pump" — and the agent returns the right hybrid crossovers, not a keyword miss.
Every lead ranked 0–100 the moment it lands, with the reasoning attached. Your team works the hot ones first instead of guessing.
Specialists don't call each other in private. They route through the orchestrator and act through MCP tools — the same open Model Context Protocol endpoint your own automation can call. Every step is logged with a confidence score.
Autonomy without guardrails is a liability in a dealership. Here's what we engineered in, on purpose — and why.
Every action carries a confidence score. Below your threshold, Miles holds the message and asks a human instead of guessing.
Seven layers — input validation, prohibited-topic and unsafe-pricing filters, after-hours rules, and output review — sit between the model and your customer.
Escalation rules hand the conversation to a rep with full context the moment a lead gets frustrated, asks for a person, or trips a rule.
Shadow mode is the standard week-one pilot: Miles drafts, your team reviews, nothing sends without a human until you flip the switch.
Miles' conversation specialists run on Claude Opus 5.5 for instruction-following and brand voice. Lead scoring uses Claude Haiku 4.5 for fast, high-volume triage. Review responses and market-research analysis use Claude Sonnet 5.5. Each job gets the right tier.
Yes — within limits you set. Confidence gating and escalation rules decide when Miles sends on its own versus when it routes to a rep. During pilots, shadow mode keeps a human on every message until you're comfortable.
It escalates. A low-confidence action, a request for a human, or a tripped guardrail hands the conversation to your team with the full thread and the reason attached — no dead ends.
The orchestrator routes each turn to a specialist, and specialists act through MCP tools — the same open endpoint your other automation can call. Every routing decision and tool call is written to an append-only audit trail with its confidence score.
Yes. Data is dealership-scoped with foreign-key isolation, PII is encrypted at rest, and conversations run through the Anthropic API rather than a shared model. You can export everything, any time.
Book a walkthrough and watch Miles run a real lead end to end — qualify, match, book — with confidence gating and shadow-mode rollout, against your current BDC workflow.