LLM, tools, retrieval, vector stores, API, databases, SaaS integrations, scheduled & event triggers — all pre-installed, zero setup.
High-risk actions (refunds, contracts, outbound emails) routed to Feishu / DingTalk / WeCom, proceed only after approval.
Duration, tokens, cost, error tracking · failure replay · version rollback · A/B testing.
Auto-retry, exponential backoff, concurrency limiting, timeout degradation — no need to build your own state machine.
API · Scheduled · Webhook · IM keyword · Email · Ticket · Folder watch · Database change.
Assign edit / run / view permissions per workflow, per node, per data source. Even outsourced personnel can collaborate safely.
Industry / scenario template library, one-click fork to your team · edit prompts and go live
New ticket → intent classification → FAQ retrieval → check order → refund (with approval) / transfer to human
Topic → outline → draft → images → layout → one-click publish to WeChat / Xiaohongshu / Weibo
New lead → company background research → BANT scoring → high score push to sales / low score send email
Inbox new resume → JD match scoring → extract candidate summary → push to BOSS / Lark
Invoice image → OCR extraction → three-way matching → reject anomalies → book normal entries
PRD submission → auto-generate OpenAPI · user stories · test cases · estimation
Use agents when the process is uncertain. The LLM decides what tool to call next, looping until the task is complete.
Use workflows when the process is defined. Humans explicitly draw each step and branch condition — fast, explainable, auditable.
WANFLOW supports both on the same canvas. You can also embed agent nodes inside workflows — combining workflow stability with agent flexibility.