AI that answers, navigates and drills into detail
AI built into both Web and the mini-program, with open APIs and an MCP server that plug the system into your workflow — usable by everyone, guarded by a seven-layer security gateway and deterministic authorization.
AI assistant (for outlet staff)
Natural language is the new interface — numbers, navigation and detail from one floating ball.
- Ask for numbers, get charts: "Customer revenue ranking last month?" "Month-to-date profit by settler?" → conclusion plus bar chart
- Navigate by describing the page — "take me to customer pricing" — with query params pre-filled
- Paste a waybill number for instant cost / revenue detail pop-ups
- Built-in manual search: new staff learn by asking instead of reading docs
- A draggable floating ball always on call; sessions starred, deleted and context-isolated
- AI usage and cost fully observable in the report center
AI customer service (for end customers)
AI first, human hand-off when needed — every inquiry is caught, and everything leaves a trail.
- Mini-program users chat with AI for instant logistics answers
- One tap escalates to outlet staff with a ticket auto-created in the service desk
- Tickets managed by status (pending / replied / closed), searchable by agent and keyword
- Agent replies from Web stream straight into the user’s mini-program chat — user, AI and agent in one thread
- Closed tickets reopen to continue the thread; the whole service trail is auditable
Enterprise-grade AI security: seven layers deep
You cannot assume a model follows rules — so the rules are enforced by deterministic code, not the model.
- Deterministic authorization: AI data scope derives solely from the auth context, resolved per request; conversation cannot widen it, and out-of-scope requests return empty sets
- Seven gateway layers: identity scope → minimal tool exposure → input guardrails → tool authorization → output field allow-lists → egress redaction → output guardrails plus audit & rate limiting
- Egress redaction: credential fields (passwords, IDs, bank cards) removed outright; personal data (phones, addresses, names) deterministically masked before reaching any model
- Default-deny, fail-closed: any uncertain layer refuses rather than passes
- Aligned with OWASP LLM Top 10, NIST AI RMF and Google SAIF
Open platform APIs: contract-grade openness
Not "here’s a doc" — a governed open platform.
- Independent open-layer authentication: dedicated API keys and tokens, fully isolated from in-system logins
- Rate limiting and scoping per key, with self-service creation and revocation inside the system
- API contracts generated from code at compile time and CI-verified for drift — docs never go stale
- Coverage across accounts, reports, waybills, exceptions and notifications — enough for a complete third-party integration
MCP & agent access: your AI, your logistics books
One API key lets your own agent ask about your logistics books.
- Standard MCP server: drop a key into Claude Desktop / Cursor and operate conversationally
- Channel & sales scenarios out of the box: open demo accounts in one sentence, on site
- Personal identity resolution: bind a WeChat account and the agent’s reach equals your own permissions — intersected, never amplified
- Plug the logistics system into enterprise systems, workflow platforms or your own AI applications via agents
- Hot-switchable LLM providers (Anthropic / Qwen and more) with platform-side usage and cost control
FAQ
What can the AI assistant answer?
Business questions for outlet staff: report figures, per-waybill accounting details and in-app navigation — ask in natural language, get answers with charts and deep links.
What is MCP integration?
Via the standard MCP protocol, system capabilities plug into AI clients such as Claude and Cursor, so authorized operations can be invoked right inside a conversation.
How is the open platform secured?
Dedicated API keys with scoped permissions, rate limiting and audit logging keep AI strictly inside its authorized boundary.
Let AI read the reports for you
Try one question on real data: "Which customer had the lowest margin last month?"
Get Started · Scan on WeChat