Leaf AI Workspace Enterprise AI Assistant

Leaf AI Workspace (leafflow-ai.vercel.app): enterprise document RAG Q&A, whitelist tools for projects / receivables, and a public demo of human confirmation before notifications.

Leaf Technology own product

What it is

A public enterprise AI assistant demo: knowledge-base Q&A, operations data lookup, and pending-confirmation workflows in one assistant UI—try without signing up.

Who it’s for

  • ・Companies adopting enterprise knowledge bases / RAG
  • ・Teams that need AI to query ERP / projects / receivables with permission control
  • ・Leads who want Agents to require human confirmation for critical actions

Background

Enterprises often want to “ask company questions like ChatGPT,” but general chat models don’t know internal documents and can’t safely read databases or send notifications. Leaf Technology built a public Workspace demo with three inspectable paths.

Problem

Wiring a single chat API can’t cite document sources, can’t query operations data, and more easily claims high-risk actions like “email sent” without human confirmation.

Solution

Policy questions retrieve indexed documents first, answer with filename and page; projects and overdue receivables use whitelist tools against a demo database; collection notices create drafts and a pending-confirmation flow—after confirm only audit is written; the public demo never sends email.

Architecture

  • ・Enterprise knowledge base / document index with citations (filename, page)
  • ・Whitelist tools: search, query projects, query overdue receivables, create drafts
  • ・Human confirmation gates and audit logs
  • ・No-signup public demo UI

Core features

  • ・Company policy Q&A with knowledge-base sources
  • ・Query in-progress projects and overdue receivables (demo data)
  • ・Collection-notice drafts and post-confirmation audit
  • ・Model cannot compose SQL—only call controlled tools
  • ・When knowledge is insufficient, clearly say so—don’t invent policy
  • ・Public demo https://leafflow-ai.vercel.app/

Technical challenges

Design “can answer,” “can query,” and “can act” separately so the model doesn’t merge query and execution into an untestable black box.

How we solved it

Documents via RAG, data via tools, actions via human confirmation; demo boundaries spell out fictional data and no real outbound email.

Results

A publicly demonstrable enterprise AI assistant architecture matching real delivery capability in RAG, API / database integration, and controlled Agents.

Visit product site →

FAQ

How is Leaf AI Workspace different from general ChatGPT?

General chat models answer from prior knowledge. Workspace searches company documents with sources, reads demo operations data via whitelist tools, and waits for human confirmation before notification-class actions.

Will AI send collection emails directly?

Not in the public demo. The system creates a draft and pending-confirmation flow; after confirm it only writes an audit log—no email is sent.

What development capability does this show?

RAG knowledge bases, controlled tool calling, human-confirmation workflows, and a publicly tryable enterprise AI product UI.

Related services

If you have a similar Web product, quiz platform, or AI productization need, start with Custom Web Systems,AI System Integration, or request a development assessment.

Have a similar product idea?

Share the target users, whether registration should be optional, languages, and scoring. We’ll assess whether a rules engine or AI integration fits better.

Free development assessment
Leaf AI Workspace Case Study | RAG Knowledge Base, Controlled Tools & Human-Confirm Agent